St. Cloud State University St. Cloud State University
theRepository at St. Cloud State theRepository at St. Cloud State
Culminating Projects in Computer Science and
Information Technology
Department of Computer Science and
Information Technology
10-2019
An Approach For Detecting Online Dating Scams An Approach For Detecting Online Dating Scams
Ozkan Kahveci
Follow this and additional works at: https://repository.stcloudstate.edu/csit_etds
Part of the Computer Sciences Commons
Recommended Citation Recommended Citation
Kahveci, Ozkan, "An Approach For Detecting Online Dating Scams" (2019).
Culminating Projects in
Computer Science and Information Technology
. 31.
https://repository.stcloudstate.edu/csit_etds/31
This Starred Paper is brought to you for free and open access by the Department of Computer Science and
Information Technology at theRepository at St. Cloud State. It has been accepted for inclusion in Culminating
Projects in Computer Science and Information Technology by an authorized administrator of theRepository at St.
Cloud State. For more information, please contact [email protected].
An Approach For Detecting Online Dating Scams
by
Ozkan Kahveci
A Starred Paper
Submitted to the Graduate Faculty of
St. Cloud State University
in Partial Fulfillment of the Requirements
for the Degree of
Master of Science
in Information Assurance
October, 2019
Starred Paper Committee:
Abdullah Abu Hussein, Chairperson
Dennis Guster
Akalanka Mailewa
2
Abstract
Online dating scam has been rapidly increasing the internet’s rapid growth
synchronically. However, there is no such tool that is available for the public to use it
and prevent online dating scams. In this paper, techniques for scam detection in
online dating websites profiles are described. A tool for automatically identifying fake
profiles on dating websites such as e-Harmony, OkCupid, match.com is used in this
paper. The web application generates a scam likelihood regarding the input profile’s
description by using the scam action components.
Regarding National Public Radio’s news recently, online dating scams had an impact
of $143 million in the United States (“Americans Lost $143 Million In Online
Relationship Scams Last Year,” 2019). This number indicates the link between the
number of users that use online dating websites and the number of scams on these
websites. The primary purpose of this paper is creating public awareness and alerting
users for whom they might be contacting online dating websites.
3
Table of Contents
Page
List of Tables ...................................................................................................................5
List of Figures...................................................................................................................6
Chapter
I. Introduction ........................................................................................................ 7
Problem Statement ..................................................................................... 10
Nature and Significance of the Problem ...................................................... 12
Objective of the Study ................................................................................. 12
Study Questions/Hypotheses ...................................................................... 12
Summary .................................................................................................... 12
II. Background and Review of Literature .............................................................. 13
Introduction ................................................................................................. 13
Background Related to the Problem ............................................................ 14
Literature Related to the Problem ............................................................... 21
Literature Related to the Methodology ........................................................ 36
Summary ..................................................................................................... 41
III. Methodology ...................................................................................................... 42
Introduction .................................................................................................. 42
Design of the Study ..................................................................................... 42
Data Collection ........................................................................................... 43
Tools and Techniques ................................................................................. 44
Summary .................................................................................................... 57
4
Chapter Page
IV. Data Presentation and Analysis ....................................................................... 58
Introduction ................................................................................................. 58
Data Presentation ....................................................................................... 58
Data Analysis .............................................................................................. 59
Summary .................................................................................................... 60
V. Results, Conclusion, and Recommendations ................................................... 62
Introduction ................................................................................................. 62
Results ........................................................................................................ 62
Conclusion .................................................................................................. 62
Future Work ................................................................................................ 63
References ............................................................................................................. 65
Appendix
A. Additional Information ...................................................................................... 69
5
List of Tables
Table Page
1. Data Samples Input Output Comparison...…………...………………...…………59
6
List of Figures
Figure Page
1. Inter-site or Cross-Site Profile Cloning (Wani & Jabin) ......................................... 30
2. Vision API ………………………………………………………………………………..38
3. Sight Engine API (SightEngine.com) ………………………………………………....39
4. Sight Engine API Sending Request (SightEngine.com) …………………………….40
5. Scam Likelihood Indicator ………………………………………………………………41
6. Implementation of Online Dating Scam Detector …………………………………….63
7
Chapter I: Introduction
With today’s online world, social networking websites not only change how
people look up for information, send or receive, but also the way people interact with
each other has been shaped and rapidly evolving in the online world (Fire, Kagan,
Elyashar, & Elovici, 2013). That is why many websites offer dating services to match
with other users on their websites based on user profile description. People on these
websites read biographies of other users and decide whether to interact or not. The
existence of reciprocity in these dating websites creates a new challenge to the public
as to whether the existing user profile is genuine or suspicious (Whitty & Buchanan,
2016).
An ever-increasing number of Americans are going to dating sites and portable
applications in order to find love and fellowship. A Pew Research Center examination
uncovered that about 60 percent of U.S. grown-ups consider internet dating a decent
method to meet individuals, and Match.com, one of the most prevalent dating
destinations, says individuals 50 and more established speak to its quickest developing
portion of clients. Be that as it may, looking for sentimental delight online can have a
noteworthy drawback: Cyberspace is brimming with scammers anxious to exploit
desolate hearts. The con works something like this: You post a dating profile and up
pops a promising match — attractive, savvy, smart, and amiable. This potential mate
professes to live in another piece of the nation or to be abroad for business or military
arrangements. Be that as it may, the person appears to be stricken and anxious to
become more acquainted with you better and recommends you move your relationship
8
to a private channel like email or a talk application. Over weeks or months, you feel
yourself developing nearer. You make arrangements to meet face to face; however, for
your new love, something consistently comes up. At that point, you get a critical
solicitation. There is a crisis (a restorative issue, maybe, or a business emergency), and
your online friend needs you to wire cash rapidly. The person will guarantee to pay it
back; however, that will never occur. Instead, the scammer will continue requesting
more until you, at long last acknowledge you have been had. Fake suitors likewise
search out focuses via web-based networking media, and they are progressively
dynamic. The Federal Trade Commission (FTC) got more than 21,300 reports of
sentiment scams in 2018, up 250 percent from three years sooner. Announced
misfortunes totaled $143 million, the most for a customer misrepresentation. The more
seasoned the person in question, the more massive the budgetary toll, as per the FTC
— the middle individual misfortune for individuals matured 70 and over was $10,000,
contrasted with $2,600 for all exploited people. The warning signs can be summarized
as it is shown below:
-Your new romantic friend sends you an image that looks progressively like a
model from a style magazine than a customary depiction.
-The individual rapidly needs to leave the dating site and speak with you through
email or texting.
-He or she lavishes you with attention. Swindlers often inundate prospective
marks with texts, emails, and phone calls to draw them in.
9
-The person in question more than once guarantees to meet you face to face yet
consistently appears to think of a reason to drop.
Last month, for instance, in the United States a man who was the victim of this
sort of trick – he related an assault procedure like that for a situation detailed in Chile in
2018 – in the wake of having met the individual through an internet dating website and
picked up his trust, the scammer mentioned the sending of cozy photographs. Soon
after they were sent, the victim got a message from a man professing to be the dad of a
minor and who took steps to document charges against him for sending a tyke an
express picture, except if he sent him two paid ahead of time 'cash cards' with US$300
each. The victim was educated that it was a trick after he had reached the police
(“When love becomes a nightmare,” 2019).
Dating services aim to make their users meet in real life and fall in love.
However, some websites cater to audiences with a specific ethnic group or cultural
background (Huang, Stringhini, & Yong, 2015). Examples are christianmingle.com,
lationoamericancupid.com, muslims4marriage.com. Some sites only focus on making
people match based on profile descriptions. These services establish a connection
between two users, and both scammers and the victim get notified. Once the
connection established, the victim is ready to expose personal information (Elovici, Fire,
& Gilad, 2015). Also, some dating services already expose user-profiles publicly, and
this is another privacy concern. Accordingly, many users publish private information,
and it gets exposed more than they thought (Elovici et al., 2015).
10
A surprising component of online romance scams is that the culprits of this kind
of crime depend on strategies intended to make victims act, of their own volition, in
manners that conflict with their advantage. To put it another way, not at all like some
other digital crimes, for example, fraud, an online romance scam is just fruitful to the
degree that a scammer can influence a victim to complete solicitations. This may help
clarify why numerous victims have announced they censure themselves for what
happened, and that, instead of getting support, they have been met with displeasure
from relatives and others (Whitty and Buchanan, 2012b). However, what might make an
individual comply with the solicitations of a scammer who might be found hundreds, or
even a great many miles away?
Analyst Robert Cialdini has distinguished six mental standards regularly abused
by those looking to pick up the compliance of others. Cialdini has talked about these
standards finally in his milestone book Influence: The Psychology of Persuasion, first
distributed in 1984 and later reexamined in 2007. Throughout the years, this book has
turned out to be required perusing in many showcasing courses and has additionally
been retooled into a school reading material, presently in its fifth version. The
compliance standards laid out by Cialdini might be a valuable system inside which to
comprehend the extraordinary impact culprits of online romance scams have over their
victims.
Intention to keep the family name alive continues from the first time of existence
of humans until now. Even though dating scams existed in the past when there was no
internet, it became so comfortable with today’s technology. When we consider why
11
scams exist in the online dating world, the financial number is the motivation behind
these scams. In 2012, the online dating market value was 1.9 billion dollars (Kopp,
Layton, Sillitoe, & Gondal, 2016). This market value attracts scammers for financial
gain. The romance scam is being fed by vulnerable people who are desperately seeking
for partners online. Also, some dating websites claim that they have over 15 million
members registered, so this is a perfect amount of population for scammers (Huang et
al., 2015). Even though simple scams such as 419 scams still exist on these websites,
advance scammers find different methods to attract emotionally vulnerable people
(Huang et al., 2015). 419 scam is in which scammers ask for money to build trust with
their victims (Huang et al., 2015).
In the United States, regarding all single users who are looking for a partner,
17.000 scams have been reported in 2017 (“Americans Lost $143 Million In Online
Relationship Scams Last Year,” 2019). Since Facebook, LinkedIn, and other social
networks offer a look-up option for existing users, dating websites only recommend
profiles for users to read (Wani & Jabin, 2017). While online dating services focus on
how to find a better way to match their users, they are not paying enough attention to
protect their users. The scam gab is getting bigger and bigger while the online dating
industry is also growing fast. Thus, improving user recommendations will potentially
help to find an eventual partner to meet in real life.
Problem Statement
Since online dating websites are not capable of verifying every user profile for
existence, a fake profile can be created easily for hostile intentions.
12
Nature and Significance of the Problem
Online dating fraud is highly successful, which causes reasonable financial and
mental damage to its victims (Whitty & Buchanan, 2016). In 2019 February NPR
reported that people lost 143 million dollars because of online dating scams
(“Americans Lost $143 Million In Online Relationship Scams Last Year,” 2019). This
study will help to improve the understanding of online dating possible scam situations.
Objective of the Study
This study will help the public to identify fake profiles on dating websites.
This study aims to create an environment where the users are more aware of possible
scams.
Study Questions/Hypotheses
Is there any software available for the public to identify the fake profiles on dating
websites?
What are the potential benefits of using this software for the users?
Hypotheses 1: There are no applications available to alert users for fake profiles on
dating websites.
Hypotheses 2: The software created by the researcher will aid the users to recognize if
the user on the other side is a scam or not.
Summary
In general, considering the previous researches, online dating security has not
been deeply researched, and this has caused both financial and mental impacts on
victims. Since the online world is expanding every other day, it brings more
13
challenges with itself. People not only look for the information they need but also look
for partners that are single and looking for relationships. This intention opens a gate
for scammers to exploit this new service as a financial gain. While the scammer uses
this hole, scammers also leave prominent scars both financially and mentally on their
target. This study focuses on how to improve the safety of online dating users against
scammers on these websites. It also tries to answer if there is such a software that
identifies fake profiles on dating websites. The hypotheses are shown below:
Hypotheses 1: There are no applications available to alert users for fake profiles on
dating websites.
Hypotheses 2: The software created by the researcher will aid the users to recognize
if the user on the other side is a scam or not.
The next chapter will focus on the background of online dating scams and existing
studies that are related to online dating services.
14
Chapter II: Background and Review of Literature
Introduction
In this chapter, background related to dating scams and its impacts on users will
be briefly explained. This brief explanation will include several reports and studies why
online dating scam is such a huge problem. The next part of this chapter will include the
studies that suggest a solution to online dating scams. Suggestions will have two
sections. The first section will be a tool, and the second section will try to explain how
this tool will be useful to prevent the online dating scam issue.
The last section in this chapter will compare the methodologies that were used
before the solution of this study. It will briefly explain if the methodology that this study
will offer was used in past studies. Then the previous methodologies will be analyzed
why they are not useful anymore with current industrial scam techniques.
Background Related to the Problem
Violations submitted utilizing PCs, and the Internet has turned out to be
omnipresent in ongoing decades. From very complex system interruptions to the most
low-tech digital stalking cases, the media is pervaded with tales about how the Internet
is utilized by offenders to take cash and data, launder cash, and carry out different
wrongdoings. A standout amongst the most worthwhile territories of cybercrime includes
online tricks, as prove by the way that, in 2015 alone, the Federal Bureau of
Investigation's (FBI) Internet Crime Complaint Center got protests totaling over $1
billion. Tricks executed over the Internet can target associations or people and may take
on a large assortment of structures. Nonetheless, of all cybercrimes answered to the
15
FBI in 2015, the second most elevated measure of misfortune was credited to a
classification of extortion that solely targets people: "Certainty Fraud/Romance" (ICCC,
2015, p. 16). Exasperatingly, proof proposes unfortunate casualties might particularly
underreport this sort of misrepresentation, and that the genuine money related effect of
this wrongdoing is far more prominent than what is shown by authority sources (Whitty
and Buchanan, 2012a).
In any case, not at all like with numerous different scams, victims of online
sentiment scams lose unmistakably more than cash. Victims of this kind of scam are
mentally maltreated, frequently for quite a long time or even years. Victims have
revealed encountering harm to connections and being left grief-stricken, humiliated, and
profoundly embarrassed (Whitty and Buchanan, 2012b). During the ongoing preliminary
of a sentiment scammer (who was eventually sentenced for duping victims of over $1.7
million), victims affirmed they had petitioned for financial protection, lost occupations,
homes, and had been exposed to extraordinary money related hardship. Past their
monetary misfortune, victims affirmed they had fallen into discouragement and, now and
again, had mulled over suicide. A few victims even affirmed that they were explicitly
mishandled, and later extorted utilizing naked photos of themselves. The irritating
subtleties of the case, which included several victims, incited the judge who directed the
case to allude to the scam as "the most destroying crime one would ever envision
without laying hands or even eyes on another individual" (DOJ, 2017, standard. 10).
The number of online dating services is increasing rapidly, and the market is expanding
more and more. However, the damage is spreading rapidly because of the lack of
16
security in the online dating world. It is creating irreparable both mental and financial
damages on its victims. In 2011, the Melbourne Herald Sun newspaper reported that
Australians lost $21 million to online dating scams (Kopp et al., 2016). In 2012, a study
showed that 230,000 individuals had been scammed by online dating services in
England (Whitty & Buchanan, 2016). This year in 2019, National Public Radio
announced that Americans lost 143$ million because of online dating scams. The
numbers indicate how a single service on the internet can have an enormous impact on
users.
The increasing numbers of scams make users question their security on dating
services. However, there has not been such a method found to solve this issue. The
complexity of online dating services is evolving every other day; however, the security of
these services is getting lower. Simple theory in cybersecurity says that complexity and
security are inversely proportional. In another saying, while the complexity of application
increases, the security of the application will go down. So, it is always important to find
the balance and keep updating the application towards that balance. Considering the
services that are available for online dating, the financial loss shows that the balance
policy has not been being applied to their services. The gap is increasing with every
other update, and unfortunately, most updates’ information shows that the application
has almost none enhancements towards security.
The lack of security in online dating platforms not only harms financially, but it
creates severe mental harms. All a sudden, being left alone by someone who declared
love for you leaves deep cuts emotionally. Studies have shown that some victims
17
experienced dominant depressive thoughts, and some even experienced suicidal
thoughts (Whitty & Buchanan, 2016). A report by the Office of Fair Trading in the United
Kingdom has shown that the scam victims lose trust, confidence, and experience
damaged self-esteem and a reduced sense of self-worth (Whitty & Buchanan, 2016).
Endeavors have been made to comprehend the advanced wonder of Internet-
based scams from various alternate points of view. One methodology has been to
attempt to decide the commonness of the problem, by either attempting to learn the
number of victims or the number of fraudulent sales. For example, a study directed by
the Federal Trade Commission (FTC) found that 10.8 percent of U.S. grown-ups had
been the victim of some fraud in 2011. In about 33% of these cases, the Internet was
referred to as the methods by which the scam was at first advanced (Anderson, 2013).
Another study, which required more than 2,300 subjects matured 40 and more
established, found that 80% of respondents had been the objective of a fraudulent offer
and that the Internet was the most significant single wellspring of these offers (Report
On FINRA, 2018).
Concerning online romance scams explicitly, the FBI has announced that, in
2015, its Internet Crimes Complaint Center got 12,509 objections identified with
"Certainty Fraud/Romance" scams (ICCC, 2015). In any case, others have
recommended information, for example, these, who are needy upon victims' self-
detailing, may significantly under mirror the true extent of the problem. For instance, a
British study directed by Whitty and Buchanan (2012a) included asking subjects
"whether they had lost cash or knew somebody by and by who had lost cash to an
18
online romance scammer" (p. 5). These discoveries, because of a delegate test of
grown-ups in the United Kingdom, have driven Whitty and Buchanan (2012a) to gauge
that more than 230,000 British natives may have been victims of online romance scams
somewhere in the range of 2008 and 2011. This number is far higher than those self-
detailed by victims to purchaser insurance organizations in the U.K. during a similar
period (Whitty and Buchanan, 2012a).
In the United States, there are various progressing endeavors to gather the
information that incorporates data identified with the fiscal misfortune continued by
victims of scams. For instance, since 1997, the FTC's Consumer Sentinel task has
gathered buyer grievances of different sorts, including those identified with scams.
Somewhere in the range of 2013 and 2015, more than four million of these grievances
identified with fraud, with shoppers announcing over $4.1 billion in misfortune. Of fraud-
related grievances, 31% of victims showed that the Internet or email was the technique
by which they were at first reached by fraudsters (Fletcher, 2019). The FBI additionally
gathers data explicitly identified with Internet-based crimes through its Internet Crime
Complaint Center. In 2015, the FBI announced that "Certainty Fraud/Romance" spoke
to the second-biggest fraud type by victim misfortune, with victims answering to have
lost over $203 million. By and large, victims of this sort of fraud announced lost $16,260
(ICCC, 2015).
One more way to deal with the study of this point has been to investigate the
conceivable hazard variables related to the victims of these sorts of scams, as well as to
endeavor to all the more likely comprehend victim defenselessness to influence
19
systems utilized by fraudsters. For instance, a study led by the AARP recognized critical
contrasts between the degrees of instruction and salary of victims of various kinds of
scams. For instance, victims of venture scams were observed bound to be school
instructed and to report a payment of $50,000 every year or more noteworthy.
Alternately, victims of lottery scams were found to have no school training, and to report
a salary of under $50,000 every year. The study additionally inspected respondents'
helplessness too many advertising style questions and found those 50 years and more
seasoned were "fundamentally progressively inspired by the influence articulations
generally speaking" (Pak and Shadel, 2011, p. 38). Concerning online romance scams,
a progression of studies led by Whitty and Buchanan (2012b) have recommended that
men (regardless of whether hetero or gay) might be almost certain than ladies to move
toward becoming victims when utilizing an internet dating webpage. Whitty and
Buchanan (2012b) have likewise noticed that victims of online romance scams will, in
general, report they became hopelessly enamored rapidly, in contradistinction to studies
exhibiting most genuine sentimental connections develop all the more gradually. This
has driven them to conjecture that numerous online romance scam victims are
"profoundly energetic to begin to look all starry eyed at, possibly leaving them helpless
against being scammed" (p. 11).
Another zone of the center has been to attempt to recognize the wellspring of
Internet-based scams: who the culprits are, and where they are physically found. Longe
and Osofisan (2011) have challenged the generally held conviction that most
development charge scams begin from West Africa, given a study of IP locations related
20
to scam messages. The study, which consolidated a sizeable level of messages named
"dating spam," found that countless scam messages started from spots outside West
Africa, for example, Europe and North America. Nonetheless, Longe and Osofisan
(2011) have recognized their study did not involve researching whether this example "is
related to the number of volumes of Africans, Asians and other settlers' moving into the
[sic] western countries" (p. 24). Others have reasoned that a superabundance of
Internet-based scams executed outside West Africa is, all things considered, executed
by West African-based scammers. For instance, Ultrascan (2014), a Dutch security firm
that gathers and breaks down information identified with development expense scams,
has noted West African culprits of development charge fraud (regularly alluded to as
"419", in reference to an area of the Nigerian reformatory code) have relocated all
through the world as of late: "There are 419 cells in about each nation on earth... 419er
tasks are on the uptick in China (both territory China and Hong Kong), and in Malaysia.
There are 419er cells in the USA, Canada, Mexico, Ghana, Brazil, Egypt, Russia, India,
Pakistan, and the Czech Republic" (Ultrascan, 2014, p. 15). Online romance scams,
which Ultrascan (2014) thinks about a variety of customary development charge fraud,
are one of the scams routinely executed by West African scammers working throughout
the world.
Due to the lack of security in online dating platforms, people have been suffering
during their journey to find happiness. Past and recent researches prove the dating
scam impact on its victims is such a big problem in the online world.
21
Literature Related to the Problem
Fire claimed that Facebook is filled with tens of millions of fake user profiles
(Fire et al., 2013). The reason behind this is that Facebook does not pay enough
attention to its users’ privacy. The proposed solution was a tool to identify fake
profiles. This tool is labeled as “Social Privacy Protector software for Facebook.” The
software has three security layers to improve security. The first layer focuses on
users’ friends' list and identifies who might pose a threat and restricts the personal
information (Fire et al., 2013). The second layer focuses on basic privacy settings,
such as what is available to the public (Fire et al., 2013). Third and the last year
focuses on third-party applications that use personal data on Facebook (Fire et al.,
2013). More than 3.000 users downloaded the tool and restricted over nine thousand
friends (Fire et al., 2013). It also removed 1,792 Facebook applications from their
user profiles (Fire et al., 2013).
The methods that were used by Fire are persuasive regarding 2013. Analyzing
the personal data, analyzing Facebook’s privacy settings, and the Facebook
applications that use personal data techniques are very accurate ways to detect fake
profiles. However, Fire stated that their applications ran into too many false-positive
flags. Private settings and the data set that was used for machine learning were not
accurately set. Thus, the accuracy of this suggested tool was not very successful in
detecting fake profiles on Facebook.
In 2014, Radford questioned the service of E-Harmony. E-Harmony is another
different type of dating service that uniquely claims itself in the industry. This service
22
states that they use science, and they have researches work for their services to make
online dating easy and convenient (Radford, 2014). These scientific statement goes as
“developed by a team of clinical experts …l is rooted in classical psychometric theory –
which uses well-established standards to measure mental abilities and traits reliably.”
(Radford, 2014). Radford thought that these services make direct and explicit claims
about the scientific validity of its matching algorithm (Radford, 2014). It was so strange
that e-Harmony did not have any reference regarding their scientific claims. Throughout
his research, Radford demanded these studies from e-Harmony. Radford found that the
website was not able to provide any research regarding their service. Later senior
research scientist Gian Gonzaga from e-Harmony claimed that the methods that e-
Harmony uses are secret and cannot be published at Society for Personality and Social
Psychology conference (Radford, 2014). He tried to offer evidence from non-peer-
reviewed studies.
Regarding Radford’s results, e-Harmony is another marketing way to get people
to register, and this is a scam. Radford calls this “Sweet Science of Scam,” and their
methodologies cannot be trusted fully without clear evidence.
Given that the casualties of online sentiment scams consent to collaborate with
the solicitations of scammers, it might be anything but complicated to embrace an
unfeeling mentality toward the individuals who have fallen prey to this sort of
wrongdoing. In any case, the way that these scams are so pervasive thus fruitful
proposes it is not right to property the activities of exploited people to ineptitude or
guilelessness. Instead, an exertion ought to be made to comprehend why the strategies
23
utilized by online sentiment scammers can make exploited people act in such
extraordinary ways. By lighting up how sentiment scammers endeavor known mental
shortcomings, it might be conceivable not just to grow better methodologies toward
alleviating this sort of wrongdoing yet likewise to advance more compassion toward
exploited people.
Cialdini has sketched out six mental rules that are frequently misused by
individuals looking to pick up the consistency of others, regardless of whether for real or
ill-conceived purposes:
Reciprocation: Individuals are bound to agree to a solicitation when they feel a
feeling of commitment or obligation toward the requestor. This feeling of commitment
might be accomplished through giving little endowments or doing apparent favors for
somebody yet may likewise be cultivated through mutual concessions: beginning with
an enormous solicitation, at that point countering with a nearly littler solicitation when
the first is rejected. This uses the inclination for individuals to feel a "commitment to
make an admission to somebody who has made an admission to us" (Cialdini, 2007, p.
37).
Commitment and Consistency: In the wake of making an underlying
responsibility, individuals regularly feel compelled to keep on acting as per that
dedication, to "legitimize [their] prior choice" (Cialdini, 2007, p. 57). Individuals will, in
general, adjust their mental self-view to responsibilities they accept they have made,
mainly when those duties are recorded, recorded, or officially made. This inclination is
24
particularly robust when an individual accepts the person has made a responsibility
"without solid outside weights" (Cialdini, 2007, p. 93).
Social Proof: Individuals are bound to think about conduct as typical and to take
part in that conduct if there is a recognition that others are doing likewise.
Liking: Individuals are increasingly inclined to consent to the solicitations of
somebody they like. Enjoying can be accomplished through any number of methods, yet
frequently incorporates engaging a physical quality, developing a feeling of
"equivalence" (individuals will in general like individuals they see as being like
themselves), compliments, nature, and building up a feeling of universal participation
toward a mutual objective.
Authority: Consistency is simpler to get when it is seen that the requestor is in a
place of power. This standard was maybe most broadly shown by Stanley Milgram's
acquiescence tests.
Scarcity: Individuals are regularly impacted to settle on choices dependent on the
dread of losing an apparent chance. The view of shortage makes a feeling of direness,
prompting blunders in judgment when individuals react by settling on choices rapidly.
This rule owes a lot of its solidarity to an idea known as mental reactance: "...whenever
a free decision is constrained or undermined, the need to hold our opportunities makes
us want them... altogether more than already" (Cialdini, 2007, p. 245).
Eloivici proposed a method for detecting spammers and fake profiles in social
networks by using supervised learning. Supervised learning can be described as
machine learning to simplify. Eloivici created different data sets regarding fake and legit
25
profiles. The invention did not target a specific social network platform. The illustrated
profiles were tested on different platforms such as Google+ and MySpace (Elovici et al.,
2015).
Eloivici set 4 different nodes to be extracted for each user. These nodes are:
a) the number of friends of the user
b) the number of communities the user is connected to
c) the number of connections between the friends of the user; and
d) the average number of friends inside each of the user’s connected
communities (Elovici et al., 2015).
Eloivici claimed that if the user has friends from different communities and the
user’s friends list contains friends less than an average user has; these can be
considered as red flags to identify the profile as fake. Later Eloivici simulation of these
techniques in different platforms with different sizes of data sets indicated that the
invention could identify fake profiles. However, Eloivici had only illustrations with limited
data sets and specific steps to identify the spammers and fake profiles in social
networks.
Huang stated that with one in five relationships in the United States starting on
one of these dating websites (Huang et al., 2015). In this manner, the attention on these
websites increased by cybercriminals and scammers. Huang had different methods to
identify scammers comparing to previous ones. First of all, Huang used large-scale data
to study instead of using illustrations or small-scale data. The methodologies that were
suggested by Huang can be categorized into four different sub-categories:
26
1) Behavioral-based Detection
2) IP Address-based Detection
3) Photograph-based Detection
4) Text-based Detection (Huang et al., 2015).
The other difference between Huang's study was the timeline. The study was
performed for eleven months to aim for better accuracy.
Huang's first methodology was focused on behaviors from scammers. While the
real people wanted to seek out for others and start a relationship, scammers had
different behaviors. This methodology analyzes the time scammers take to respond or
send the first messages, and also it analyzes the number of conversations that are held
by scammers simultaneously (Huang et al., 2015). The behavioral system mechanism
shows similarity to the anti-spam system that is proposed by the research community
(Huang et al., 2015).
The second methodology that was proposed by Huang was an IP address-based
detection system. If a scammer created different profiles by using the same network,
this indicates serious attention as a red flag for the detection system (Huang et al.,
2015).
The third methodology is focused on the profile photos that are used by
scammers. The studies show that scammers use the same photo over and over (Huang
et al., 2015). These photos usually are an attractive young woman or a handsome
middle-aged man (Huang et al., 2015). The detection system deploys a system that
detects duplicated photos and flags those as malicious activity (Huang et al., 2015).
27
The last method is called a text-based detection system. This system checks if
the same message was sent to different users over and over. This is actually how spam
detectors work, and Hu describes that scammers use the same method also in online
dating websites.
Hu analyzed the application market for dating applications in China. Since the
Google play store is not available in some parts of China, the different application
markets were also included in the research. Over 2.5 million applications in the
application market, 967 dating applications were found for dating purposes (Hu et al.,
2018). The analysis shows that significant numbers of people downloaded these
applications. Later Hu dived into the applications they found and analyzed it. Hu
presently endeavor to distinguish fake accounts from another perspective, i.e., the
collaboration patterns. On the off chance that the accounts are genuine people, at that
point, the messages ought to be significant to the subject of the discussion. In this
manner, Hu play out a field concentrate on investigating the communication patterns of
these fraudulent dating apps. For every family, Hu arbitrarily pick an application and
introduce it on a genuine gadget. At that point, Hu register two accounts (1 male client
and one female client) to sign in and begin a discussion. Moreover, we buy the premium
administration for each application and analyze the outcomes when obtaining their
administrations.
Regarding China’s population, a total of 967 applications was downloaded over a
billion times (Hu et al., 2018). Hu and his colleagues also analyzed the reviews on these
applications for each different application market, and the results were also showing
28
that most of these applications that had over 100k reviews and most reviews were fake.
Not only the reviews were fake, later Hu selected random applications among these 967
dating applications. The signatures on these applications were mostly similar, and that
showed that the same developer distributed these applications into different markets
with different names and different company titles (Hu et al., 2018). Hu's researches also
identified that the applications were mostly targeting fake young beautiful women
profiles. The responses were not related to the topic and some applications on the
market required to purchase VIP or premium membership to respond to the messages
(Hu et al., 2018). Hu purchased the premium membership to take a further analysis
step. The total cost of 22 memberships was 176 US dollars. The research identified that
once the premium was purchased, the victim was able to respond; however, the bots
stop responding. Therefore, the entire point on these applications was to convince
victims to purchase. The accumulated revenue estimated for these applications that
were conducted in research was around 200 million US Dollars to 2 billion US dollars.
Dating websites are closely related to fake profiles on social networks such as
Twitter, Facebook, and Instagram. So, Wani dives into different methods to identify fake
profiles on social networks. Wani also compares the cloned profiles and fake profiles
and their different intentions on scam levels. Wani gives a thorough characterization of
various real and ghost profiles with an accentuation on informal online organizations.
Compromised accounts are in reality official accounts; however, their proprietors do not
have extensive oversight over these, and they have lost the control to a phisher or any
malware specialist. As per an investigation, compromised accounts are the most
29
troublesome sort of accounts to be identified. Another ongoing examination says over
97% of profiles are compromised instead of fake. The fake profiles are, for the most
part, made to take the accreditations from genuine clients, and after that, fake profiles
are deserted or deactivated.
Compromised profiles have much worth since they have officially settled a
dimension of trust inside their system and in this manner, cannot be adequately
recognized and expelled by the specialist co-ops. Assailants, for the most part, use
compromised profiles with critical consideration to use the dimension of trust. The
writers have exhibited a way to deal with distinguishing compromised accounts from two
mainstream online person to person communication destinations, Facebook and
Twitter, by recognizing profiles that show unexpected changes in the conduct by
utilizing measurable demonstrating and peculiarity recognition. Facebook has a
framework to recuperate hacked accounts once announced. There is an alternative "my
account was hacked" on the Facebook help page. One more examination uncovers that
the compromised genuine profiles spread more noxious substances than different kinds
of fake profiles (Wani & Jabin, 2017). Profile cloning is the burglary of personality from a
current user's profile and to make another fake profile utilizing stolen accreditations
(Wani & Jabin, 2017). We can say that profile cloning is the way toward taking the
injured individual's private data to make one more profile that can secure the private
data of unfortunate casualty's companions. These assaults are called Identity Clone
Attacks (ICAs), which are of two sorts of profile cloning assaults to be specific single site
and cross-site profile cloning. The scammers are generally very much financed, talented
30
people and have nearly everything accessible available to them and have authority over
bargained and contaminated records (Wani & Jabin, 2017).
Fake profiles are not quite the same as cloned ones from various perspectives. If
there should be an occurrence of cloned profiles, an enemy makes one more profile of
the effectively existing one, which is not the situation for fake profiles. Cloned profiles
are, for the most part, made to extricate the data of an injured individual or his/her
companions through the fake profiles are utilized for different purposes like spamming,
promoting, and so forth. A few people make fake profiles to have one more account,
while some make numerous accounts intentionally to go into individuals' sub-charts.
There are two different ways to make fake profiles: one is made by composing content,
and another is by physically making one more account.
What is more, there are three fundamental purposes behind making fake profiles:
First, Online Social Network specialist co-ops permit one account for every versatile
association or per email-id, and to defeat this farthest point, individuals make one more
account utilizing diverse email-ids or telephone numbers. The second is to improve the
Figure 1: Intersite or Cross-Site Profile Cloning (Wani & Jabin, 2017)
31
ubiquity or the dimension of trust among the others. The third is to spread spam content
among the genuine users. Fake elements exist wherever on the web like long range
interpersonal communication websites, shopping sites, exchange sites and gatherings,
web-based dating websites, banking frameworks, and so on. Furthermore, there is still a
need to fortify safety efforts being utilized by online informal community sites to lessen
editing fake profiles and to keep away from their dangers on interpersonal
organizations. Fake profiles are destructive for OSNs and can be progressively
hazardous in the future if not distinguished at the beginning period.
A bot is a computer program that delivers a few information to connect with
people, particularly the people utilizing the web (netizens) to change their conduct. Bots
produce over 60% of the complete web information. Online bots otherwise called web
robots or basically, a bot is a computer program that performs different assignments
rapidly and consequently, which were impractical for a human alone. Necessarily the
bots were intended to help the people to accelerate their work and make it programmed.
The first job of bots was to naturally total substance from different news sources,
function as a programmed responder to customer inquiries, go about as a therapeutic
master to determine wellbeing related issues, and programmed travel control. Be that
as it may, these days, the bots are abused by the general population in different spaces.
In informal organizations, bots are utilized to retweet a post without checking its source
to make it viral. In online multiplayer diversions, bots are utilized to pick up the uncalled
for the preferred position. Now and then, bots go about as computerized symbols to
interface with people and make informal communities, which are much progressively
32
hard to distinguish. Bots can likewise be utilized to impact users, presenting messages,
and on send companion demands in informal online organizations.
Additionally, the gathering of individuals/association who are probably going to get
influenced by the interruption of these personalities were likewise referenced in the
table. From the practical perspective, bots are comparative as Sybil accounts, yet the
principle contrast is that Sybil accounts are taken care of by users physically while bots
are mechanized computer programs. The principle utilization of bots is web information
slithering where a straightforward online computer program recognizes and separates
the data from web servers at a lot higher speed, which was unrealistic by a human
alone. Bots intended for malevolent exercises have turned into a genuine risk for the
web. Different OSN specialist co-ops utilized a few different ways to battle the
spambots. For instance, Twitter and Facebook have included an alternative "report as
spam" to recognize a spam bot.
Facebook additionally has its Facebook Immune System (FIS) to manage such
issues. At the same time, the exploration in this area is in its beginning times. Users in
different OSNs guarantee that the discovery systems are getting their official accounts.
As per an examination, over 8% of bots exist in Twitter organize. The more significant
part of them has been created for business purposes. Bots can be of two sorts
generous and harmful.
Police in the southern area of Guangdong, China have busted large rings of web-
based dating scammers who acted like appealing ladies to cheat men into purchasing
expensive tea and different items, Guangzhou Daily reported (2018). Nearby police
33
reported at the preparation that they had broken 13 packs and caught 1,310 suspects
(Sixth Tone, 2018). Each pack could approach up to 1,500 unfortunate casualties for
every month, as their individuals would lure numerous men all the while. "These scam
packs [succeed] because they catch the brain science of numerous men: When
confronting lovely ladies, men lose their judgment, and feel too timid even to consider
refusing," a Guangdong cop said at the preparation. As per the police, the possess
utilized models' photos and the People Nearby capacity on WeChat, China's most
common informing application, to bait unfortunate casualties into visiting. Following
quite a while of structure up trust, they would cheat their "beaus" by utilizing anecdotal
family disasters as stratagems to advance tea, wine, or interest in valuable metals
through a stage constrained by the gathering. One long con included homegrown tea.
Police clarified that the groups had built up a 60-day recipe for extortion: The scam
specialists would go through 15 days calmly visiting, five days pushing the relationship
into a sentiment, and 20 days demonstrating the beau that they had gone to deal with a
weak granddad in the place where they grew up, where they were likewise figuring out
how to deliver tea. The pack even enlisted models in hot jeans to posture for
photographs and recordings in southeastern China's Yunnan region, one of the nation's
outstanding tea-delivering areas. The scam would achieve its peak over the most recent
20 days when the beau would be more than once influenced to purchase costly tea
leaves. When the objective acknowledged he had been had, he would get blocked. As
indicated by Guangzhou Daily, the vast majority of the scalawags were recent college
grads, while others were "moderately aged uncles who got a kick out of the chance to
34
pick at their feet." Some systems sold tea, while others utilized tobacco, liquor,
wellbeing items, or even oil depictions as their lure of decision. Many individuals have
been captured for partaking in web-based dating scams in China. In 2017, police in
eastern China's Zhejiang area busted a task that had utilized photographs of lovely
ladies to draw exploited people into lotteries constrained by their ring. Another case in
northeastern China's Liaoning region in 2016 saw 500 suspects captured.
Here is a real online dating scam story based in Australia. (Commission, 2015)
Georgina’s children signed her up to Facebook and gave her some basic lessons on
how to use it.
‘They told me everyone was using it and that it would help us keep in touch and
see photos of my grandchildren.’ One day Georgina received a friend request from a
serviceman on peacekeeping duties in Afghanistan. She decided to accept the request
and allowed 'Jim' to be her Facebook friend. It did not start as a romance, but he said
he was lonely and looking for friends to keep him company while he was stuck on duty
in the middle of nowhere. Soon after befriending her, Jim told Georgina he had lost his
wife to cancer, and his story of looking after her was similar to her own experience when
her husband had died of cancer. ‘He then said he was being posted to Nigeria, but his
time in the U.S military was nearly finished. He sent me pictures which I now know were
stolen from someone on the internet. He kept saying he could not wait for us to be
together. We became very close, and he emailed me every day, saying it was easier for
him than using Facebook.’ Jim, who was a scammer, told Georgina he liked gemstones
and wanted to set up a jewelry store when he retired. He said this was the best part of
35
being in Nigeria because it was close to where the precious stones were being mined,
and he could buy them very cheaply. He told Georgina he was coming to see her but
had some trouble with his bank card not working in Nigeria and could not get funds to
pay for an export tax on his gemstones. Georgina transferred some money to him to
cover the tax, which he explained was only two percent of the value of the gemstones
but still amounted to $15 000. It was much money to send, but she figured he was a
right and honest serviceman, and if things worked out, they would spend the rest of their
lives together. ‘All was going well until his stopover in Malaysia. Customs officials seized
the gemstones and demanded payment to have them released. This time they were
asking $20 000. I told him it would take some time to get the money, and I had to
borrow against the family home.’ Georgina sent the money to Malaysian officials but
was told Jim was now in jail for smuggling and that she needed to contact his lawyer.
‘The lawyer said he needed to get an Anti-terrorism and Money Laundering certificate,
and this would be another $10 000. He said he also needed to pay for Jim’s court costs
plus his fees, and this would be another $5000.’ Georgina sent the money, but then Jim
said there was another government official demanding payment to extend his visa while
he waited for the court to process all the documents. ‘Almost every day, I was contacted
with a new demand for money. They sent me certificates signed by officials, forms to fill
out, and bills for everything. If you wanted to get anything done quickly, you had to pay
another fee. It seemed to me that the whole Malaysian government was corrupt. I do not
know exactly how much money I sent, but it was well over $100 000. I did not care
about money. I just wanted to help Jim, and I honestly thought he would pay me back.’
36
Even when Georgina ran out of money, the demands didn’t stop. Unsure of what to do,
Georgina finally talked to the police. They explained that her experience included the
characteristic features of a dating and romance scam, and it would be implausible she
would get her money back. She cannot help feeling in her heart that she let Jim down,
but she knows that it was all a scam.
Literature Related to the Methodology
Scam techniques on dating websites have been evolving while there is no such
free tool that is available to prevent their impacts on society. The methodology that will
offer the solution for this issue has two techniques to identify possible scam activity.
The first technique will focus on user profiles’ description and look for keywords
that are previously identified based on past researches. This methodology will be called
keyword matching. Whitty (2016) identified that if a user indicates that he is in the
military or outside the country, that is a red flag for scam activity. Also, NPR indicates
that if a user is acting like he or she is in love with the person they have not met in real
life, it is a red flag. Not common grammar errors or errors in idioms will also indicate
such a scam activity. The text-based methodology will focus on these matters and will
add the total red flags in the scam likelihood.
The second methodology will scan for profile photos. If a photo that is used in the
profile is public, this will indicate that the user profile is not real and designed for scam
activities. The tool will compare the public database and the user profile and will
generate a result. Scammers usually create a profile by using a handsome male or
female public photo. The public profile that is identified on a profile will add another level
37
into scam likelihood. This method will use different approaches to achieve the goal. The
first approach will be using Vision API. Then as a second approach, Vision API will be
integrated with Cloud Auto ML Vision. As the third approach, Vision API, Cloud AutoML
Vision will be integrated with Google Cloud Platform (GCP). Vision API is powered with
Google’s deep learning models and provides:
-Face and landmark detection
-Explicit content detection
-Label detection
-Optical Character Recognition (OCR)
AutoML gives the option to train the application for custom label detection using
Google's Neural Architecture Search and state-of-the-art transfer learning.
38
Figure 2: Vision API (Google)
Another image scanner API is called sightengine. This API was created with over
40.000 references to detect images that are commonly used in romance scams. The
scam discovery motor works by perceiving appearances of individuals known to be
utilized in scammer profiles. So regardless of whether a given picture of such an
39
individual is new and obscure to our database, we will almost certainly hail it. The API
was developed to detect nudity, celebrities, minors, image quality, artificial texts,
weapons, alcohol, offensive & hate signs, face detection, scammers, and faces hidden
with sunglasses. Here is an example of the use of sightengine API for detecting
celebrities in profile photos.
Figure 3: Sight Engine API (SightEngine.com)
Code in plain text:
// if you have not already, install the SDK with "npm install sight engine --save"
var sightengine = require('sightengine')('{api_user}', '{api_secret}');
sightengine.check([‘{model}’]).set_url('https://sightengine.com/assets/img/examples/exa
mple7.jpg').then(function(result) {
// The API response (result)
}).catch(function(err) {
// Handle error
});
40
Another example of the use of API for scammers:
Figure 4: Sight Engine API Sending Request (SightEngine.com)
Code in plain text:
// if you haven't already, install the SDK with "npm install sightengine --save"
var sightengine = require('sightengine')('{api_user}', '{api_secret}');
sightengine.check(['scam']).set_url('https://sightengine.com/assets/img/examples/examp
le7.jpg').then(function(result) {
// The API response (result)
}).catch(function(err) {
// Handle error
});
SightEngine Scam API gives an option to send 2000 requests per month at no
cost.
In conclusion, once user copies and the pastes the profile link in the dating
website into the tool, the tool will generate a scam likelihood with a graph (Figure 5).
41
The tool will use the red flags that are generated by the key-word scam technique and
public photo scanner technique.
Figure 5: Scam Likelihood Indicator
Summary
This chapter covered two main topics. First, background work was covered.
Background work included both psychological work and software work that has been
done so far. Previous researches that tried to solve or analyze the scam on dating web
services were also introduced. The background section was categorized based on the
publish years by the researches. It was set as old to newer. The second section
introduced the idea of the tool. Later sections discussed what could be used to make
this tool alive. Two main APIs that were discussed how they would fit into this idea. The
way of using their libraries was also introduced with a few simple lines of coding.
The next chapter will be more related to how the tool will be designed and used.
The methodologies for the tool will be explained at a deeper level.
42
Chapter III: Methodology
Introduction
This part will clarify what sort of study framework with the upsides and downsides
to offering an answer for the online dating scam issue. It will profoundly clarify the AI
calculation that is offered by Amazon to extract information from pictures in rekognition
API. Likewise, the advantages and disadvantages of rekognition API will be clarified in
detail. Later the exploration will incorporate outcome information to demonstrate how it
very well may be utilized to take care of the issue.
Design of the Study
With everything taken into account, both qualitative and quantitative research
plans offer different understandings reliant on research focuses. This paper includes the
obstructions and characteristics of both research plans. All through the past
examinations exhibit that if there is a remarkable piece of total data that exists and
consistent assurances supporting the data, inquiries about should move towards
quantitative research plan. In any case, as to the examination point, there might be
inadequate intelligent data or proof exhibited. Thus, the analyst will need to put a
solitary effort to demonstrate what ought to be done to comprehend the issue. This sort
of research subject will require a qualitative research plan. To plot, examine structures
may fluctuate or can be united reliant on the exploration subject and past examinations.
Since the presence of past examinations are deficient in taking care of the issue,
and the past investigations' information is not sufficient for the tackling issue, this
investigation will utilize the mixed strategy to take care of the issue.
43
Data Collection
The techniques that were utilized by Fire are extremely persuading with respect
to 2013. Investigating the individual information, breaking down Facebook's security
settings, and the Facebook applications that utilization individual information systems
are exact approaches to recognize the phony profiles. In any case, Fire expressed that
their applications kept running into such a large number of false-positive banners.
Interior settings and the informational collection that was utilized for AI were not
precisely set. Along these lines, the exactness of this recommended device was not
fruitful to distinguish the phony profiles on Facebook.
Eloivici guaranteed that if the client has companions from various networks and
the client's companion’s rundown contains companions, not exactly a regular client has;
these can be considered as warnings to recognize the profile as phony. Later Eloivici's
(2015) recreation of these strategies in various stages with various sizes of
informational indexes demonstrated that the creation could distinguish phony profiles. In
any case, Eloivici had just outlined with constrained informational collections and
specific means to distinguish the spammers and phony profiles on interpersonal
organizations.
Past researches were showing information for various zones that are sub
identified with online dating scam issues. Because of this reality, more research should
have been finished. This examination will hold a light into this issue.
44
Tools and Techniques
Amazon Rekognition makes it simple to add a picture and video examinations to
your applications. You give a picture or video to the Amazon Rekognition API, and the
administration can distinguish objects, individuals, content, scenes, and exercises. It
can distinguish any improper substance too. Amazon Rekognition additionally gives
profoundly precise facial examination and facial acknowledgment. You can distinguish,
dissect, and analyze faces for a wide assortment of utilization cases, including client
confirmation, classifying individuals checking, and open wellbeing.
Amazon Rekognition depends on the equivalent demonstrated, exceptionally
versatile, profound learning innovation created by Amazon's PC vision researchers to
dissect billions of pictures and recordings day by day. It requires no AI skill to utilize.
Amazon Rekognition incorporates a basic, simple to-utilize API that can rapidly break
down any picture or video record that is put away in Amazon S3. Amazon Rekognition
is consistently gaining from new information, and we continually include new names and
facial acknowledgment highlights to the administration. Typical use cases for utilizing
Amazon Rekognition incorporate the accompanying:
• Searchable picture and video libraries – Amazon Rekognition make pictures
and put away recordings accessible so you can find items and scenes that show up
inside them.
• Face-based client check – Amazon Rekognition empowers your applications to
affirm client characters by contrasting their live picture and a reference picture.
45
• Sentiment and statistic investigation – Amazon Rekognition translate
enthusiastic articulations, for example, glad, dismal, or shock, and statistic data, for
example, sexual orientation from facial pictures. Amazon Rekognition can break down
pictures, and send the feeling and statistic ascribes to Amazon Redshift for intermittent
giving an account of patterns, for example, in-store areas and comparable situations.
Note that a forecast of enthusiastic demeanor depends on the physical appearance of
an individual's face as it were. It is not demonstrative of an individual's inner
enthusiastic state, and Rekognition ought not to be utilized to make such an assurance.
• Facial acknowledgment – With Amazon Rekognition, you can scan for pictures,
put away recordings, and spilling recordings for appearances that match those put away
in a holder known as a face accumulation. A face accumulation is a record of
countenances that you claim and oversee. Recognizing individuals dependent on their
appearances requires two remarkable strides in Amazon Rekognition:
1. Record the countenances.
2. Search the countenances.
•Unsafe substance discovery – Amazon Rekognition can recognize grown-up
and natural substances in pictures and put away recordings. Engineers can utilize the
returned metadata to channel unseemly substance dependent on their business needs.
The past is hailing a picture dependent on the nearness of dangerous substance, and
the API additionally restores a various leveled rundown of marks with certainty scores.
These names demonstrate explicit classes of a dangerous substance, which empowers
granular sifting and the board of vast volumes of client created content (UGC). Models
46
incorporate social and dating locales, photograph sharing stages, websites and
discussions, applications for kids, web-based business destinations, stimulation, and
internet promoting administrations.
• Celebrity acknowledgment – Amazon Rekognition can perceive VIPs inside
provided pictures and in recordings. Amazon Rekognition can perceive a large number
of big names over a few classifications, for example, governmental issues, sports,
business, diversion, and media.
• Text recognition – Amazon Rekognition Text in Image empowers you to
perceive and separate literary substance from pictures. Content in Image bolsters most
text styles, including exceptionally adapted ones. It distinguishes content and numbers
in various directions, for example, those usually found in standards and publications. In
picture sharing and internet-based life applications, you can utilize it to empower visual
hunt dependent on a list of pictures that contain similar watchwords. In media and
excitement applications, you can index recordings dependent on the meaningful content
on the screen, for example, advertisements, news, sports scores, and subtitles. At last,
in open wellbeing applications, you can recognize vehicles dependent on tag numbers
from pictures taken by road cameras.
Recognize celebrities. The celebrity API returns a variety of famous people
perceived in the info picture. RecognizeCelebrities restores the 100 most prominent
faces in the picture. It records perceived VIPs in the CelebrityFaces exhibit and
unrecognized faces in the UnrecognizedFaces cluster. RecognizeCelebrities doesn't
return big names whose appearances aren't among the most prominent 100 faces in the
47
picture. For every superstar perceived, RecognizeCelebrities restores a Celebrity
object. The Celebrity item contains the VIP name, ID, URL connects to extra data,
coordinate certainty, and a ComparedFace object that you can use to find the VIP's face
on the picture. Amazon Rekognition doesn't hold data about which pictures a big name
has been perceived in. Your application must store this data and utilize the Celebrity ID
property as one of a kind identifier for the big name. In the event that you don't store the
big name or new data URLs returned by RecognizeCelebrities, you will require the ID to
distinguish the superstar in a call to the GetCelebrityInfo activity. You pass the info
picture either as base64-encoded picture bytes or as a kind of perspective to a picture
in an Amazon S3 can. On the off chance that you utilize the AWS CLI to call Amazon
Rekognition tasks, passing picture bytes isn't upheld. The picture must be either a PNG
or JPEG designed document.
For instance, see Recognizing Celebrities in an Image:
This activity expects authorization to play out the rekognition:
RecognizeCelebrities activity.
Solicitation Syntax
{
"Picture": {
"Bytes": mass, "S3Object": {
"Basin": "string", "Name": "string", "Rendition": "string"
}
}
48
Solicitation parameters. The solicitation acknowledges the accompanying
information in JSON design.
The information picture as base64-encoded bytes or an S3 object. In the event
that you utilize the AWS CLI to call Amazon Rekognition activities, passing base64-
encoded picture bytes isn't bolstered.
On the off chance that you are utilizing an AWS SDK to call Amazon
Rekognition, you won't have to base64-encode picture bytes passed utilizing the Bytes
field.
Reaction Syntax
{
"CelebrityFaces": [
{
"Face": {
"BoundingBox": { "Tallness": number, "Left": number, "Top": number, "Width":
number
},
"Certainty": number, "Tourist spots": [
{
"Type": "string", "X": number, "Y": number
} ],
"Posture": {
"Pitch": number, "Move": number, "Yaw": number
49
},
"Quality": {
"Splendor": number,
"Sharpness": number }
},
"Id": "string", "MatchConfidence": number, "Name": "string",
"Urls": [ "string" ]
} ],
"OrientationCorrection": "string", "UnrecognizedFaces": [
{
"BoundingBox": {
"Stature": number, "Left": number, "Top": number, "Width": number
},
"Certainty": number, "Tourist spots": [
{
"Type": "string", "X": number, "Y": number
} ],
"Posture": {
"Pitch": number, "Move": number, "Yaw": number
},
"Quality": {
"Splendor": number,
50
"Sharpness": number }
} ]
356
Amazon Rekognition Developer Guide RecognizeCelebrities
}
Reaction elements. On the off chance that the activity is productive, the
administration sends back an HTTP 200 reaction. The accompanying information is
returned in JSON design by the administration.
Insights concerning every big-name found in the picture Amazon Rekognition can
distinguish a limit of 15 big names in a picture.
Type: Array of Celebrity objects OrientationCorrection
The direction of the info picture (counterclockwise course). On the off chance that
your application shows the picture, you can utilize this incentive to address the direction.
The jumping box directions returned in CelebrityFaces, and UnrecognizedFaces speak
to confront areas before the picture direction is remedied.
On the off chance that the information picture is in .jpeg group, it may contain
interchangeable picture (Exif) metadata that incorporates the picture's direction.
Assuming this is the case, and the Exif metadata for the information picture populates
the direction field, the estimation of OrientationCorrection is invalid. The CelebrityFaces
and UnrecognizedFaces jumping box directions speak to confront areas after Exif
metadata is utilized to address the picture direction. Pictures in .png configuration don't
contain Exif metadata.
51
Type: String
Legitimate Values: ROTATE_0 | ROTATE_90 | ROTATE_180 | ROTATE_270
Unrecognized faces. Insights concerning each unrecognized face in the picture.
Type: Array of ComparedFace objects
Mistakes
Access denied exception. You are not approved to play out the activity. HTTP
Status Code: 400
Image too large exception. The info picture size surpasses as far as possible.
For more data, see Limits in Amazon Rekognition
HTTP Status Code: 400
Internal server error. Amazon Rekognition encountered an assistance issue.
Attempt your call once more. HTTP Status Code: 500
Amazon Rekognition Developer Guide RecognizeCelebrities
Invalid image format exception. The gave picture organization isn't bolstered.
HTTP Status Code: 400
InvalidImage format exception. The gave picture organization isn't upheld.
HTTP Status Code: 400
Invalid parameter exception. The information parameter damaged a
requirement. Approve your parameter before calling the API activity once more.
HTTP Status Code: 400
Invalid s3 object exception. Amazon Rekognition can't get to the S3 article
determined in the solicitation. HTTP Status Code: 400
52
Provisioned throughput exceeded exception.The number of solicitations
surpassed your throughput limit. In the event that you need to expand this point of
confinement, contact Amazon Rekognition.
HTTP Status Code: 400
Throttling exception. Amazon Rekognition is briefly unfit to process the
solicitation. Attempt your call once more. HTTP Status Code: 500
Search faces. For a piece of the given information, face ID scans for
coordinating countenances in the accumulation the face has a place with. You get a
face ID when you add a face to the accumulation utilizing the IndexFaces activity. The
activity thinks about the highlights of the information face with countenances in the
predefined accumulation.
We can likewise look through appearances without ordering faces by utilizing the
SearchFacesByImage activity.
The activity reaction restores a variety of countenances that match, requested by
likeness score with the most noteworthy similitude first. All the more explicitly, it is a
variety of metadata for each face coordinate that is found. Along with the metadata, the
reaction additionally incorporates certainty esteem for each face coordinate,
demonstrating the certainty that the particular face coordinates the information face.
This activity expects authorization to play out the rekognition: SearchFaces
activity.
The solicitation acknowledges the accompanying information in the JSON group.
The ID of the accumulation the face has a place with.
53
Type: String
Length Constraints: Minimum length of 1. The most extreme length of 255.
Example: [a-zA-Z0-9_.\-]+
Required: Yes
FaceId
The ID of a face to discover matches for in the accumulation.
Type: String
Example: [0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}
Required: Yes
Face match threshold. Discretionary worth indicating the base trust in the face
match to return. For instance, don't restore any matches where trust in matches is under
70%. The default worth is 80%.
Type: Float
Substantial Range: Minimum estimation of 0. Most final estimation of 100.
{
"CollectionId": "string", "FaceId": "string", "FaceMatchThreshold": number,
"MaxFaces": number
}
Required: No
MaxFaces
The most extreme number of appearances return the data. The activity restores
the most extreme number of countenances with the most elevated trust in the match.
54
Type: Integer
Substantial Range: Minimum estimation of 1. Greatest estimation of 4096.
Required: No
Reaction Syntax
{
"FaceMatches": [
{
"Face": {
"BoundingBox": { "Stature": number, "Left": number, "Top": number, "Width":
number
},
"Certainty": number, "ExternalImageId": "string", "FaceId": "string", "ImageId":
"string"
},
"Closeness": number }
],
"FaceModelVersion": "string", "SearchedFaceId": "string"
}
Reaction Elements
On the off chance that the activity is sufficient, the administration sends back an
HTTP 200 reaction. The accompanying information is returned in the JSON position by
the administration. FaceMatches
55
A variety of countenances that coordinated the info face, alongside the trust in
the match. Type: Array of FaceMatch objects
FaceModelVersion
Adaptation number of the face location model related to the info gathering
(CollectionId). Type: String
SearchedFaceId
The ID of the face that was looked for matches in a gathering.
Type: String
Example: [0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}
Access denied exception. You are not approved to play out the activity. HTTP
Status Code: 400
inner server blunder
Amazon Rekognition encountered an assistance issue. Attempt your call once
more. HTTP Status Code: 500
Invalid parameter exception. The information parameter disregarded a
limitation. Approve your parameter before calling the API activity once more.
HTTP Status Code: 400
Provisioned throughput exceeded exception. The number of solicitations
surpassed your throughput limit. On the off chance that you need to expand this cutoff,
contact Amazon Rekognition.
HTTP Status Code: 400
56
Resource not found exception. The accumulation indicated in the solicitation
can't be found. HTTP Status Code: 400
Throttling exception. Amazon Rekognition is incidentally incapable of
processing the solicitation. Attempt your call once more. HTTP Status Code: 500
The logic can be simplified, as shown below for the solution design.
Figure 6: Implementation of Online Dating Scam Detector
57
Summary
This chapter dived into the solution that was implemented by using Amazon
Rekognition API and Selenium library in .NET. The details of what Amazon Rekognition
API offers were explained in detail. After APIs capabilities were explained, the exception
handling was also explained in detail by referring to the Amazon API documentation on
Amazon.com. At the end of the chapter, the solution design diagram was introduced.
58
Chapter IV: Data Presentation and Analysis
Introduction
The primary objective of Amazon Rekognition API distinguishes the celebrity face
in the given picture. In this part, I will attempt a trial to see better how the API would
create an outcome with a similar celebrity input yet in various circumstances. Every one
of the information that was captured was available publicly. A significant piece of
information was not exhibited since the objective was to make a place that may create
issue results. Toward the finish of this trial, it will give how trustworthy the API that was
utilized to illuminate the online dating scam issue.
Data Presentation
This area will demonstrate example information to test Amazon Rekognition API.
The primary segment will display the name of the individual's picture that was submitted
into API, and the subsequent section will report the outcome that was created by
artificial intelligence. All the photographs that were sustained into API were copied from
google pictures. Some male celebrities' photos were acquainted with API with facial hair
and no facial hair. Likewise, some female celebrities' photographs were submitted to
API with short and long hair. A portion of the photos that are not celebrities were
additionally nourished into API to perceive what results were produced. The explanation
was an information test was made to see how exact outcomes were created by Amazon
Rekognition artificial intelligence API.
59
Data Analysis
Table 1
Data Samples Input Output Comparison
Input Image
Rekognition API Result
Jenna Fischer
Jenna Fischer
Angelina Jolie
Angelina Jolie
George Clooney (Short Facial Hair)
George Clooney
George Clooney (Long Facial Hair)
No Celebrity Faces Recognized
Abu Abdullah Hussein
No Celebrity Faces Recognized
Dennis Guster
Enrique Krauze
Mailewa Akalanka
No Celebrity Faces Recognized
Michelle Obama
Michelle Obama
Madonna
Madonna
Madonna (Wearing an Eye Patch)
Porcelain Black
Lebron James (Facial Hair)
Lebron James
Adele
Adele
Adele (Short Hair)
Adele
Rainn Wilson (No Facial Hair)
Rainn Wilson
Rainn Wilson (Facial Hair)
No Celebrity Faces Recognized
60
Test results in details:
- Amazon Rekognition API was fed with a picture in which Madonna was wearing an
Eye Patch. The result was a celebrity; however, it was Porcelain Black. This was a
false-positive result for Amazon Rekognition.
- Amazon Rekognition API was fed with a picture in which Rainn Wilson had facial hair.
API was not able to detect the celebrity. It was detected that if a celebrity had facial hair,
API was not able to detect the celebrity. The result was indicated as ‘No celebrity was
detected.’
-Dennis Guster is a professor at St. Cloud State University in Information Assurance
Department. His picture was downloaded from the department’s website and was used
as an input for Amazon Rekognition API. The result was surprising, and it detected the
professor’s picture as a celebrity. The indicated result was Enrique Krauze who is a
Mexican historian.
Summary
Information that was fed into Amazon Rekognition API shows that celebrities who
are wearing some stuff that covers their appearances may bring about an inappropriate
yield. In like manner, stars who are known without having facial hairs in motion pictures
are likewise making some bogus yield results as the example information has
demonstrated that a non-celebrity face can be perceived as a celebrity in certain
circumstances, which is another off-base outcome. In any case, later, the information
61
that was encouraged into API with celebrities' photographs, which are appeared with a
similar facial appearance in the movies they acted, was adequately perceived by
Rekognition API.
62
Chapter V: Results, Conclusion, and Recommendations
Introduction
This section clarifies the aftereffects of the offered solution. The parts that the
appropriate response was fruitful in actualizing and the result of what was accomplished
was defined. The techniques and the libraries that were utilized during the execution
were portrayed in detail. Later it plunges into future work, and the client situations
should be possible.
Results
The software that was implemented was able to detect the celebrity photos on
dating websites. During the trial version, the software was set for the OkCupid dating
website. By using the selenium front-end automation, the HTML tags in the OkCupid
site was automated to find the user profile photos. The profile was set with a celebrity
photo which was Steve Carell. Amazon Rekognition API was able to identify the
celebrity and displayed a notification with the celebrity name. However, different
celebrities from outside the US was put into the profile, Amazon Rekognition API had
shown a different name for the celebrity, even though it was able to identify the photo as
a celebrity.
Conclusion
The examination showed that online profiles could be automated dependent on
client contribution with Selenium. When the info is gotten, a picture can be filtered with
Amazon Rekognition to comprehend whether the profile has a celebrity name
photograph or not.
63
Future Work
Over time artificial intelligence will improve itself, and facial recognition will serve
better results. For future work, researchers can focus on how to automate better on
dating profiles yet not only the websites. This research was mainly focused on data
scraping on websites and using artificial intelligence to detect faces. However,
applications on phones have been trendy rather than websites. Today’s world people
use their phones way often than computers for dating purposes. Hence, a future
researcher should make an app both on android and apple markets. The same idea can
be improved by using more scam flags to identify for better results. Also, this idea can
be integrated with dating platforms such as Tinder’s or Match.com’s apps on the apps
market and can be served to the public under Tinder Safe+ or Match+.
In addition, instead of automating through the front-end, the same idea for the
future can be automated through back end once dating services provide their public
API. Data scraping through the back end will be way faster, and it will undoubtedly
create fewer errors and better exception handling. This will also bring a better user
experience.
There were more indicators detected as scam flags through this research.
However, all the indicators were not implemented in the application. For future
researches, more scam indicators can be implemented to serve better to the public. For
example, if the IP address of the profile can be detected, and it shows that the IP
address is not from the U.S, this can be implemented. Furthermore, if the dating profile
64
had any text that was implying, he or she is outside the U.S, the profile intro can be
scanned and added to the implementation.
As Facebook also came out with their dating environment, the expectation is it
will become more popular, and since there is more research on scams on Facebook,
there can be another application build specifically for Facebook for the future. Like the
fact that Facebook has the top number of users compared to any other social network
environment, the research will help millions out there by only explicitly focusing on
Facebook.
65
References
Americans Lost $143 Million In Online Relationship Scams Last Year. (2019, Feb).
Retrieved June 8, 2019, from NPR.org website: https://www.npr.org/2019/02
/13/694171341/americans-lost-143-million-in-online-relationship-scams-last-year
AWS Global Infrastructure. (2019). Machine Learning in the AWS Cloud. doi:
10.1002/9781119556749.ch7
Cialdini, R. B. (2007). Influence: the psychology of persuasion. New York:
HarperCollins.
DOJ Grants Financial Guide (2017). Retrieved from:
https://ojp.gov/financialguide/doj/pdfs/DOJ_FinancialGuide.pdf.
Elovici, Y., Fire, M., & Gilad, K. (2015). Method For Detecting Spammers and Fake
Profiles in Social Networks.
Fletcher, E. (2019, February 14). Romance scams rank number one on total reported
losses. Retrieved from https://www.ftc.gov/news-events/blogs/data-
spotlight/2019/02/romance-scams-rank-number-one-total-reported-losses.
Fire, M., Kagan, D., Elyashar, A., & Elovici, Y. (2013). Friend or Foe? Fake Profile
Identification in Online Social Networks. ArXiv:1303.3751 [Physics]. Retrieved
from http://arxiv.org/abs/1303.3751
Hu, Y., Wang, H., Zhou, Y., Guo, Y., Li, L., Luo, B., & Xu, F. (2018). Dating with
Scambots: Understanding the Ecosystem of Fraudulent Dating Applications.
ArXiv:1807.04901 [Cs]. Retrieved from http://arxiv.org/abs/1807.04901
66
Huang, J., Stringhini, G., & Yong, P. (2015). Quit Playing Games with My Heart:
Understanding Online Dating Scams. In M. Almgren, V. Gulisano, & F. Maggi
(Eds.), Detection of Intrusions and Malware, and Vulnerability Assessment (Vol.
9148, pp. 216–236). https://doi.org/10.1007/978-3-319-20550-2_12
Kopp, C., Layton, R., Sillitoe, J., & Gondal, I. (2016). The Role of Love stories in
Romance Scams: A Qualitative Analysis of Fraudulent Profiles. International
Journal of Cyber Criminology, 9(2), 205–217.
https://doi.org/10.5281/zenodo.56227
Longe, O., & Osofisan, A. (2011). On the Origins of Advance Fee Fraud Electronic
Mails: A Technical Investigation Using Internet Protocol Address Tracers.
Retrieved from https://digitalcommons.kennesaw.edu/ajis/vol3/iss1/2/
Pak, K., & Shadel, D. (2011). https://assets.aarp.org/rgcenter/econ/fraud-victims-11.pdf.
Retrieved from https://assets.aarp.org/rgcenter/econ/fraud-victims-11.pdf
Radford, B. (2014). The sweet science of seduction or scam? Evaluating eHarmony.
Skeptical Inquirer, 38(6), 38-. Retrieved from Expanded Academic ASAP.
Sixth Tone. (2018, June 1). Over 1,300 Arrested in 'Tea Leaves' Online Dating Scam.
Retrieved from https://www.sixthtone.com/news/1002397/over-1,300-arrested-in-
tea-leaves-online-dating-scam.
2018 Report on FINRA Examination Findings. (2018, December 7). Retrieved from
https://www.finra.org/rules-guidance/guidance/reports/2018-report-exam-findings.
67
2015 Internet Crime Report. (2015). Retrieved from
https://pdf.ic3.gov/2015_IC3Report.pdf.
Ultrascan Advanced Global Investigations. (2014). Retrieved from
https://www.ultrascan-agi.com/public_html/html/pdf_files/Pre-Release-
419_Advance_Fee_Fraud_Statistics_2013-July-10-2014-NOT-FINAL-1.pdf.
Wani, M. A., & Jabin, S. (2017, May). A sneak into the Devil’s Colony- Fake Profiles in
Online Social Networks. 31.
What You Need to Know About Romance Scams. (2019, September 10). Retrieved
from https://www.consumer.ftc.gov/articles/what-you-need-know-about-romance-
scams.
When love becomes a nightmare: Online dating scams. (2019, February 14). Retrieved
June 9, 2019, from WeLiveSecurity website:
https://www.welivesecurity.com/2019/02/14/love-becomes-nightmare-scams-
apps-online-dating-sites/
Whitty, M. T., & Buchanan, T. (2016). The online dating romance scam: The
psychological impact on victims – both financial and non-financial. Criminology &
Criminal Justice, 16(2), 176–194. https://doi.org/10.1177/1748895815603773
Vision AI | Derive Image Insights via ML | Cloud Vision API | Google Cloud. (2019,
Aug). Retrieved from
https://cloud.google.com/vision/?utm_source=google&utm_medium=cpc&utm_ca
68
mpaign=na-US-all-en-dr-skws-all-all-trial-b-dr-1003905&utm_content=text-ad-
none-any-DEV_c-CRE_291263994208-
ADGP_Hybrid+|+AW+SEM+|+SKWS+|+US+|+en+|+BMM+~+ML/AI+~+Vision+A
PI+~+Vision+Api-KWID_43700036256019077-kwd-
475110369966&utm_term=KW_+vision +api-
ST_+Vision++Api&gclid=Cj0KCQjwivbsBRDsARIsADyISJ_Tx87_0XVvub0EsVw
QER0DuyGNVulhJh9j7e2w93iTd7GF02rvxGgaAvz1EALw_wcB.
69
Appendix A: Additional Information
Here is the code for the Online Scam Detector Tool:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Web;
using System.Web.UI;
using System.Web.UI.WebControls;
using OpenQA.Selenium;
using OpenQA.Selenium.Chrome;
namespace WebApplication1
{
public partial class Main : Page
{
public static ChromeOptions options = new ChromeOptions();
public static ChromeDriver driver = new ChromeDriver(options);
protected void Page_Load(object sender, EventArgs e)
{
}
protected void Unnamed1_Click(object sender, EventArgs e)
{
//Setting up chrome profile
70
options.AddArguments(@"user-data-dir=" + "C:\\Users\\ramiz\\OneDrive\\Desktop\\dist"
+ "\\profile");
//Loading the Same Profile For Login Purposes
options.AddArguments(@"user-date-dir" +
"C:\\Users\\ramiz\\AppData\\Local\\Google\\Chrome\\User Data");
options.AddArguments("--no-startup-window");
//Login
driver.Navigate().GoToUrl("https://www.okcupid.com/login");
IWebElement email = driver.FindElement(By.CssSelector("#root > span > div > div >
div.login2017-container > span > div > form > div.login2017-fields > div:nth-child(1) >
span.oknf-typable-wrapper.oknf-typable-wrapper--text > input"));
//Accessing email value from the html form and sending input
email.SendKeys(Request.Form["email"]);
//Send Password
IWebElement password = driver.FindElement(By.CssSelector("#root > span > div > div
> div.login2017-container > span > div > form > div.login2017-fields > div:nth-child(2) >
span.oknf-typable-wrapper.oknf-typable-wrapper--password > input"));
password.SendKeys(Request.Form["pass"]);
System.Threading.Thread.Sleep(2500);
//Click On Login
IWebElement login = driver.FindElement(By.CssSelector("#root > span > div > div >
div.login2017-container > span > div > form > div.login2017-actions > input"));
71
login.Click();
//Wait for the automation process
System.Threading.Thread.Sleep(2500);
//Validate Login Session
try
{
IWebElement userExist = driver.FindElement(By.CssSelector("#navigation > div >
span:nth-child(2) > div.profile-button-container > button > div > div > div"));
Label1.Text = "Login Success!";
Response.Redirect("http://localhost:63581/Url");
}
catch
{
Label1.Text = "Login Failed!";
}
}
}
}
Url.aspx
using System;
using System.Collections.Generic;
using System.Linq;
72
using System.Web;
using System.Web.UI;
using System.Net;
using System.Web.UI.WebControls;
using OpenQA.Selenium;
using OpenQA.Selenium.Chrome;
using Amazon.Rekognition;
using Amazon.Rekognition.Model;
using System.Drawing;
using System.IO;
namespace WebApplication1
{
public partial class Url : System.Web.UI.Page
{
private string inputUrl;
protected void Page_Load(object sender, EventArgs e)
{
if(Page.IsPostBack)
{
inputUrl = TextBox1.Text;
}
}
73
public void Analyze_Click(object sender, EventArgs e)
{
//Navigate to the input Profile to Analyze
Main.driver.Navigate().GoToUrl(inputUrl);
//Click on the profile thumbnail pic
IWebElement profilepic = Main.driver.FindElement(By.ClassName("profile-thumb"));
profilepic.Click();
//Download the profile Image
ITakesScreenshot ssdriver = Main.driver as ITakesScreenshot;
Screenshot screenshot = ssdriver.GetScreenshot();
Screenshot tempImage = screenshot;
//Saving the image to analyze
tempImage.SaveAsFile(@"C:\Users\ramiz\source\repos\WebApplication1\WebApplicati
on1\ProfilePics\image.png");
string photo =
@"C:\Users\ramiz\source\repos\WebApplication1\WebApplication1\ProfilePics\image.pn
g";
//Rekognition API
AmazonRekognitionClient rekognitionClient = new AmazonRekognitionClient();
RecognizeCelebritiesRequest recognizeCelebritiesRequest = new
RecognizeCelebritiesRequest();
Amazon.Rekognition.Model.Image img = new Amazon.Rekognition.Model.Image();
74
byte[] data = null;
try
{
using (FileStream fs = new FileStream(photo, FileMode.Open, FileAccess.Read))
{
data = new byte[fs.Length];
fs.Read(data, 0, (int)fs.Length);
}
}
catch (Exception)
{
WarningLabel.Text = ("Failed to load file " + photo);
return;
}
img.Bytes = new MemoryStream(data);
recognizeCelebritiesRequest.Image = img;
WarningLabel.Text=("Looking for celebrities in image " + photo + "\n");
RecognizeCelebritiesResponse recognizeCelebritiesResponse =
rekognitionClient.RecognizeCelebrities(recognizeCelebritiesRequest);
WarningLabel.Text=(recognizeCelebritiesResponse.CelebrityFaces.Count + "
celebrity(s) were recognized.\n");
foreach (Celebrity celebrity in recognizeCelebritiesResponse.CelebrityFaces)
75
{
WarningLabel.Text=("This profile is using a celebrity photo: " + celebrity.Name);
}
}
protected void Cancel_Click(object sender, EventArgs e)
{
//Cleanup the text box
TextBox1.Text = "";
}
}
}
Here is the front-end code:
<%@ Page Title="Main" Language="C#" MasterPageFile="~/Site.Master"
AutoEventWireup="true" CodeBehind="Main.aspx.cs" Inherits="WebApplication1.Main"
%>
<asp:Content ID="BodyContent" ContentPlaceHolderID="MainContent" runat="server">
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!--
==================================================================
76
=============================-->
<link rel="icon" type="image/png" href="Images/icons/favicon.ico"/>
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css" href="Vendor/bootstrap/css/bootstrap.min.css">
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css" href="fonts/font-awesome-4.7.0/css/font-
awesome.min.css">
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css" href="fonts/iconic/css/material-design-iconic-
font.min.css">
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css" href="Vendor/animate/animate.css">
<!--
==================================================================
77
=============================-->
<link rel="stylesheet" type="text/css" href="Vendor/css-
hamburgers/hamburgers.min.css">
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css" href="Vendor/animsition/css/animsition.min.css">
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css" href="Vendor/select2/select2.min.css">
<!--
==================================================================
=============================-->
<link rel="stylesheet" type="text/css"
href="Vendor/daterangepicker/daterangepicker.css">
<!--
==================================================================
=============================-->
<link href="CSS/css/main.css" type="text/css" rel="stylesheet" />
<link href="CSS/css/util.css" type="text/css" rel="stylesheet" />
<!--
78
==================================================================
=============================-->
</head>
<style>
#services {
width: 100%;
border:0px;
outline:0px;
}
</style>
<body>
<div class="limiter">
<div class="container-login100">
<div class="wrap-login100">
<form class="login100-form validate-form">
<span class="login100-form-title p-b-26">
Welcome To Safer Dating
</span>
<span class="login100-form-title p-b-48">
<i class="zmdi zmdi-font"></i>
<asp:Label ID="Label1" runat="server" Text=""></asp:Label>
</span>
79
<div class="wrap-input100 validate-input" data-validate = "Valid email is: [email protected]">
<input class="input100" type="text" name="email">
<span class="focus-input100" data-placeholder="Email"></span>
</div>
<div class="wrap-input100 validate-input" data-validate = "Please select service">
<input class="input100" type="text" name="dating">
<select name="services" id="services">
<option value="okcupid">OkCupid</option>
<option value="match">Match.com</option>
<option value="eharmony">EHarmony</option>
<option value="tinder">Tinder</option>
</select>
<span class="focus-input100" data-placeholder="Dating Service"></span>
</div>
<div class="wrap-input100 validate-input" data-validate="Enter password">
<span class="btn-show-pass">
<i class="zmdi zmdi-eye"></i>
</span>
<input class="input100" type="password" name="pass">
<span class="focus-input100" data-placeholder="Password"></span>
</div>
<Asp:Button class="login100-form-btn" runat="server" Text="Login"
80
OnClick="Unnamed1_Click" Height="34px" Width="300px" />
<!--
==================================================================
=============================-->
<script src="Vendor/jquery/jquery-3.2.1.min.js"></script>
<!--
==================================================================
=============================-->
<script src="Vendor/animsition/js/animsition.min.js"></script>
<!--
==================================================================
=============================-->
<script src="Vendor/bootstrap/js/popper.js"></script>
<script src="Vendor/bootstrap/js/bootstrap.min.js"></script>
<!--
==================================================================
=============================-->
<script src="Vendor/select2/select2.min.js"></script>
<!--
==================================================================
=============================-->
<script src="Vendor/daterangepicker/moment.min.js"></script>
81
<script src="Vendor/daterangepicker/daterangepicker.js"></script>
<!--
==================================================================
=============================-->
<script src="Vendor/countdowntime/countdowntime.js"></script>
<!--
==================================================================
=============================-->
<script src="Scripts/main.js"></script>
</body>
</html>
</div>
</div>
</div>
</asp:Content>
<!-- Url Aspx Page -->
<%@ Page Language="C#" AutoEventWireup="true" CodeBehind="Url.aspx.cs"
Inherits="WebApplication1.Url" %>
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
<head runat="server">
<link rel="stylesheet" type="text/css" href="CSS/cssButton/button.css">
82
</head>
<body>
<form id="form1" runat="server">
<div>
<p> Please enter the url of suspected profile.</p>
<asp:TextBox ID="TextBox1" class="advancedSearchTextBox"
runat="server"></asp:TextBox>
</div>
<button style="--content: 'Analyze';" type="submit" name="btnAnalyze" id="btnAnalyze"
runat="server" onserverclick="Analyze_Click">
<div class="left"></div>
Analyze
<div class="right"></div>
</button>
<button style="--content: 'Cancel';" type="button" name="btnCancel" id="btnCancel"
runat="server" onserverclick="Cancel_Click">
<div class="left">
</div>
Cancel
<div class="right"></div>
</button>
</br>
83
</br>
</br>
<asp:Label ID="WarningLabel" runat="server" Text=""></asp:Label>
</form>
</body>
</html>
