=Paper=
{{Paper
|id=Vol-2667/paper22
|storemode=property
|title=Analysis of open data of a social network in order to identify deviant communities
|pdfUrl=https://ceur-ws.org/Vol-2667/paper22.pdf
|volume=Vol-2667
|authors=Rostislav Mikherskii,Dmitry Kuznetsov
}}
==Analysis of open data of a social network in order to identify deviant communities ==
https://ceur-ws.org/Vol-2667/paper22.pdf
Analysis of open data of a social network in order to
identify deviant communities
Rostislav Mikherskii Dmitry Kuznetsov
Physico-technical Institute Physico-technical Institute
V.I. Vernadsky Crimean Federal University V.I. Vernadsky Crimean Federal University
Simferopol, Russia Simferopol, Russia
mrm03@mail.ru dimabrayankuznetsov@mail.ru
Abstract—The system of analysis of open data of the social emergence of group norms, and the use of linguistic markers.
network Vkontakte is developed and programmatically Similar studies have been conducted for groups promoting
implemented. Two ways of identification of deviant suicidal behavior [15].
communities are proposed. The first way is by the number of In [16-20], the categorization of pornographic content
community subscribers blocked by the social network for
violating the rules. The second way, by the presence of
and the frequency of its use were studied.
common subscribers between the studied community, and the In [21], the authors also focused on consumption
community about which it is precisely known that it is deviant. networks for adult content, which is present in many online
It is experimentally established that the second method of social networks and on the Internet as a whole. The authors
identification of deviant communities gives the best result. of this work investigated how such communities interact
with the entire social network. They found that few small and
Keywords—big data, open data, social network closely related communities are responsible for much of the
I. INTRODUCTION production of content. Produced content is distributed
through the rest of the network mainly directly or through
Analysis of open data from social networks is a bridge communities, reaching at least 450 times more users.
significant area in the field of big data processing. In In this work, a demographic analysis of the networks of
particular, an important task for both law enforcement producers and consumers of adult content was also carried
agencies and social network administrators is to identify out. It has been shown that it is possible to easily identify
communities of these networks that disseminate socially several key users in order to radically eradicate the process
dangerous content.Many works that were written recently of distribution of pornographic content.
have been devoted to discussion of this problem.The work The issue of community polarization in social networks
[1] is devoted to the development of a method for assessing was studied in detail in [22]. It proposed a new polarization
the degree of connectedness of user profiles of social metric based on the analysis of the boundary of a pair of
networks based on open data. The degree of connectedness (potentially polarized) communities, which better reflects the
of user profiles is understood as the probability of meeting concepts of antagonism and polarization.
profile owners in real life.In [2], a review of methods that Cyber aggression, as a form of deviant behavior in the
detect the demographic attributes of a user from their profile Internet environment, was studied in detail in [23-28]. This
and messages is made. In [3,4], forms of deviant behavior of socio-psychological phenomenon has many forms, the main
users of the Russian-language segment of the Internet are of which are trolling, cybermobbing and astroturfing.
examined in detail.In particular, in [4] it was shown that the As can be seen from the above review of published
main reason for deviant behavior in social networks is scientific papers, the search for deviant communities is an
virtuality and anonymity. In [5], according to foreign important task both for scientists involved in researching
sources, a review of the main methods of analysis of social such communities and for law enforcement agencies.
networks in relation to the task of identifying suspicious and Unfortunately, most often, the identification of deviant
criminal communities is carried out. communities is carried out manually, often only by user
To study social networks in terms of social relationships, complaints.
the Social Network Analysis (SNA) method is often used. The aim of this work was to develop a methodology for
The SNA method is described in detail in [6–8]. In this identifying deviant communities in the social network
method, the objects of research are the nodes, and the Vkontakte in automatic mode. To achieve this goal, two
relationships characterizing the relationship between them. options have been proposed to search for such communities.
Nodes can be communities, users of social networks, etc.
The connections between these nodes can be money II. RESULTS
transfers, communication, friendship, etc. This method has In the first version, the following algorithm for searching
been successfully used to study the organization structure of for such communities is proposed and programmatically
the Al-Qaida terrorist network [9], to study the network of implemented. For the studied community, the number l of
terrorist organizations operating in India [10], to analyze the subscribers blocked by the social network for breaking the
topological structure of criminal networks, in particular the rules, as well as the total number L of subscribers of this
network of methamphetamine traffic [11]. These research
studies are mainly motivated by the need to find effective community, is determined. The coefficient k l is found. It
L
methods to undermine criminal or terrorist organizations.
Anorexia-oriented online communities have been studied is assumed that if the coefficient k is greater than some
in [12–14]. A wide range of issues was studied in these critical value kd, then the community under study is deviant.
works, including the construction and management of The software implementation of the above algorithm was
member identities, the processes of social recognition, the implemented in the Python programming language. During
Copyright © 2020 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0)
Data Science
the implementation of this program, 50,704 communities of deviant community is determined. It is assumed that a
the Vkontakte social network were randomly selected. In sufficiently large number of communities from this list will
order to shorten an influence of statistical error, only also be deviant. This algorithm was programmatically
communities with 100 or more total subscribers were implemented using the Python programming language.
selected from the general list. Due to system and API To test the performance of this program, the deviant
limitation, communities with a few members were community “Mom Anarchy” was chosen with an
considered. A coefficient was calculated for each of these identification number of 177615404. This community is
communities. Further, all communities were sorted in engaged in popularizing the ideas of anarchism and has
descending order of magnitude of this coefficient. Table 1 32097 subscribers. The data processing time was 18 hours.
presents the first 20 communities from the list. Followers of this community are also subscribed to 940512
other communities. All of them were sorted in descending
TABLE I. COMMUNITIES WITH A HIGH PERCENTAGE OF BLOCKED order by the number of users who are also subscribed to the
SUBSCRIBERS
Mama Anarchy community. Table 2 presents the first 20
№ Community Number of Number of Percentage of communities from this list.
identification subscribers blocked blocked
number in the subscribers subscribers
TABLE II. COMMUNITIES WHOSE SUBSCRIBERS ARE ALSO
community from the total SUBSCRIBERS OF THE MOM ANARCHY COMMUNITY
number of
community № Community Number of The number
subscribers, k• identification number subscribers in the of
100% community subscribers
1 172017411 104 101 97.1154 who are also
2 171896750 122 114 93.4426 subscribers
3 41398959 107 98 91.5888 of the Mama
4 125043269 1017 904 88.8889 Anarchy
5 19613748 960 852 88.75 community
6 176328754 226 193 85.3982 1 *** 5539982 15035
7 148023353 495 419 84.6465 2 91050183 9356399 12924
8 188941498 530 438 82.6415 3 *** 707327 12712
9 23811356 1116 921 82.5269 4 159146575 1162785 12521
10 164252296 152 123 80.9211 5 *** 563784 11987
11 150230769 198 157 79.2929 6 *** 4403183 11644
12 130381011 200 157 78.5 7 *** 2768306 11317
13 154988787 410 317 77.3171 8 *** 2508543 11246
14 155397881 847 654 77.2137 9 57846937 11275065 11224*
15 149830913 107 81 75.7009 10 *** 2684988 11154
16 170030633 577 428 74.1768 11 *** 2586853 11145
17 *** 174 129 74.1379 12 150550417 937052 10916
18 164288533 153 113 73.8562 13 149094324 2076903 10832*
19 143657800 424 312 73.5849 14 30316056 1809325 10451
20 157513161 420 309 73.5714 15 66678575 4976245 10299
16 12353330 3555825 10167
17 154168174 1264550 10145
In order to prevent propaganda of deviant communities,
18 173556111 641480 10005
in this table and further in table 2, the identification number 19 *** 3802683 9576
of all such communities is replaced by the symbols “***”. 20 133180305 3116645 9540
As can be seen from this list, there is only one deviant
community in it (community under No. 17). This community As can be seen from this table, out of 20 communities of
was classified as deviant due to the presence of pornographic the presented list, 9 are deviant. The main reasons that these
material in it. communities are attributed to deviants are: propaganda of
Thus, the hypothesis that the percentage of blocked users violence, criticism of the existing constitutional system, and
in deviant communities is greater than in non-deviant the use of profanity. This results show us that algorithm
communities has not been experimentally confirmed. results should not be considered as final predictions but as an
Furthermore, it’s clear that some communities are abandoned assumption. Still results must be managed by special person
and they can contain a lot of banned users because of lack of to make a conclusion about community content. The main
moderation and new subscribers. Other communities can be aim of this algorithm for now is to narrow the search for
related to advertising or temporary events. But they are still deviant communities.
not deviant despite the fact that social network Vkontakte has Furthermore, this algorithm allows to consider
special rules that restrict the creation of such communities as communities with bigger amount of subscribers. However, it
communities with an inappropriate content. should be mentioned that API has a strong impact on
The second option for searching for deviant communities algorithms productivity. Therefore such systems have a
is based on the following algorithm: One community is portability limitations. Nevertheless, the core idea of this
found for which it is known for certain that it is deviant. For system is to show dependencies between blocked users
this community, a list of subscribers is defined. Each of these amount and community content.
subscribers defines the communities to which it is The scientific novelty of this work lies in the proposed
subscribed. For each of the communities in this list, the algorithm, which helps to identify deviant communities.
number of subscribers who are also subscribers of the studied Despite the fact that current algorithm can only help us to
VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 99
Data Science
make suggestion about community content, there could be [5] M. Basarab, I. Ivanov, A. Kolesnikov and V. Matveev, “Detection of
illegal activities in cyberspace based on the analysis of social
ways to improve it by using extra algorithms and tools, such networks: algorithms, methods and tools (review),” Cybersecurity
as image recognition tools and text analyzer. Therefore, issues, vol. 4, no. 17, pp. 11-19, 2016.
holistic recognition system could be developed to make more [6] L.C. Freeman, “The development of social network analysis: A study
accurate predictions about deviant communities in social in the sociology of science,” Social Networks, vol. 27, no. 4, pp. 377-
networks with open API. 384 , 2005.
[7] A. Hopkins, “Graph theory, social networks and counter terrorism,”
III. CONCLUSION Univ. of Massachusetts Dartmouth, pp. 22, 2010.
[8] L.C. Freeman, “Centrality in social networks conceptual
Thus, the second way of identifying deviant communities clarification,” Social networks, vol. 1, no. 3, pp. 215-239, 1979.
is much more effective than the first. This technique for [9] V. Krebs, “Mapping networks of terrorist cells,” Connections, vol. 24,
identifying deviant communities in automatic mode can be no. 3, pp. 43-52, 2002.
applied not only on the social network Vkontakte but also in [10] P. Choudhary and U. Singh, “A survey on social network analysis for
other social networks. We also note that the second method counter-terrorism,” Int. Journal of Computer Applications, vol. 112,
no. 9, pp. 24-29, 2015.
can be applied not only to search for deviant communities,
[11] J. Xu, H. Chen, “The topology of dark networks,” Communications of
but also when searching for communities related to the the ACM, vol. 51, no. 10, pp. 58-65, 2008.
studied community, for example, in marketing research. In [12] J. Gavin, K. Rodham and H. Poyer, “The presentation of “pro-
the case of such studies, it is possible to determine the anorexia” in online group interactions,” Qualitative Health Research,
interests of community users and, accordingly, build a policy vol. 18, no. 3, pp. 325-333, 2008.
to attract new users to this community. [13] J.D.S. Ramos, A.D.F PereiraNeto and M. Bagrichevsky, “Pro-
Another possible use of this method is to conduct an anorexia cultural identity: characteristics of a lifestyle in a virtual
community,” Interface (Botucatu), vol. 15, no. 37, pp. 447-460, 2011.
advertising campaign of a certain community. In this case, as
[14] N. Boero and C.J. Pascoe, “Pro-anorexia communities and online
the studied community, you can choose the community interaction: Bringing the pro-ana body online,” Body & Society, vol.
whose advertising you want to conduct. Define a list of 18, no. 2, pp. 27-57, 2012.
communities associated with this community and place [15] S.M. Haas, M.E. Irr, N.A. Jennings and L.M. Wagner,
advertising messages in these communities. “Communicating thin: A grounded model of Online Negative
Enabling Support Groups in the pro-anorexia movement,” New Media
It should also be noted that to search for deviant & Society, vol. 13, no. 1, pp. 40-57, 2010.
communities, it may be useful to use machine learning [16] M. Schuhmacher, C. Zirn and J. Volker, “Exploring youporn
methods, such as, for example, artificial immune systems categories, tags, and nicknames for pleasant recommendations,”
[29-31] or convolutional neural networks [32-41]. However, Workshop on Search and Exploration of X-Rated Information. ACM,
even when using machine learning, the method of identifying pp. 27-28, 2013.
deviant communities by the presence of common subscribers [17] G. Tyson, Y. Elkhatib, N. Sastry and S. Uhlig, “Are People Really
Social in Porn 2.0?” Proceedings of the 9th International AAAI
between the studied community, and the community about Conference on Web and Social Media (ICWSM), pp. 436-444, 2015.
which it is known for certain that it is deviant will not lose its [18] G.M. Hald and A. Stulhofer, “What types of pornography do people
relevance. This is primarily due to the fact that this method use and do they cluster? Assessing types and categories of
has a high degree of transparency in interpreting the results pornography consumption in a large-scale online sample,” Journal of
obtained, in contrast to machine learning methods, which are Sex Research, pp. 1-11, 2015.
often a black box, the results of which are often [19] G.M. Hald, N.N. Malamuth, T. Lange, “Pornography and sexist
attitudes among heterosexuals,” Journal of Communication, vol. 63,
incomprehensible. no. 4, pp. 638-660, 2013.
Thus, in this study, a new method is proposed that allows [20] G.M. Hald, “Gender differences in pornography consumption among
you to quickly, cheaply and efficiently search for deviant young heterosexual danish adults,” Archives of sexual behavior, vol.
communities. 35, no. 5, pp. 577-585, 2006.
[21] M. Coletto, L.M. Aiello, C. Lucchese and F. Silvestri, “On the
ACKNOWLEDGMENT Behaviour of Deviant Communities in Online Social Networks,”
Proceedings of the 10th International AAAI Conference on Web and
In conclusion, we would like to thank Marina Social Media (ICWSM), pp. 72-81, 2016.
Vsevolodovna Glumova, Director of the Physico-technical [22] P.H.C. Guerra, Jr.W. Meira, C. Cardie, R. Kleinberg, “A measure of
Institute of the V. I. Vernadsky Crimean Federal University, polarization on social media networks based on community
and Victor Vasilyevich Milyukov, head of the Department of boundaries,” Proceedings of the 7th International AAAI Conference
on Web and Social Media (ICWSM), pp. 215-224, 2013.
computer engineering and modeling of the Physico-technical
Institute of the V. I. Vernadsky Crimean Federal University, [23] J.S. Chibbaro, “School counselors and the cyberbully: interventions
and implications,” Journal of Professional School Counseling, vol.
for their assistance in organizing research. 11, no. 1, pp. 65-68, 2007.
[24] R. Gable, J. Snakenborg and R. Van Acker, “Cyberbullying:
REFERENCES Prevention and Intervention to Protect Our Children and Youth,”
[1] V. Kataeva, I. Pantyukhin and I. Yurin, “Methods for assessing the Preventing School Failure, vol. 55, no. 2, pp. 88-95, 2011.
degree of connectivity of social network user profiles based on open [25] W. Heirman and M. Walrave, “Cyberbullying: Predicting
data,” Open education, vol. 21, no. 6, pp. 14-22, 2017. Victimisation and Perpetration,” Children & Society, vol. 25, pp. 59-
[2] A. Gomzin and S. Kuznetsov, “Methods for constructing socio- 72, 2011.
demographic profiles of Internet users,” Proceedings of the ISP RAS, [26] J.S. Donath, “Identity and Deception in the Virtual Community,”
vol. 27, no. 4, pp. 129-142, 2015. Communities in Cyberspace,” London: Routledge, pp. 26, 1999.
[3] A. Baklantseva, “Transformation of social norms and deviations in [27] N.E. Willard, “From Cyberbullying and Cyberthreats: Responding to
the Russian-language Internet,” News of universities in the North the Challenge of Online Social Aggression, Threats, and Distress,”
Caucasus region. Social sciences, vol. 3, pp. 21-25, 2014. Champaign, IL: Research Press, pp. 303, 2007.
[4] D. Cherenkov, “Deviant behavior in social networks: causes, forms, [28] R.A. Vnebrachnykh, “Trolling as a form of social aggression in
consequence,” Nauka-Rastudent.Ru, vol. 7, no. 19, pp. 29, 2015. virtual communities,” Bulletin of the Udmurt University. Philosophy.
Sociology. Psychology. Pedagogy, vol. 1, pp. 48-51, 2012.
VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 100
Data Science
[29] R. Mikherskii, “Application of an artificial immune system for visual [36] C. Lian, M. Liu, J. Zhang and D. Shen, "Hierarchical Fully
pattern recognition,” Computer Optics, vol. 42, no. 1, pp. 113-117, Convolutional Network for Joint Atrophy Localization and
2018. DOI: 10.18287/2412-6179-2018-42-1-113-117. Alzheimer's Disease Diagnosis Using Structural MRI," IEEE
[30] G. Luh, “Face recognition based on artificial immune networks and Transactions on Pattern Analysis & Machine Intelligence, vol. 42, no.
principal component analysis with single training image per person,” 04, pp. 880-893, 2020. DOI: 10.1109/TPAMI.2019.2895781.
Immune Computation, vol. 2, no. 1, pp. 21-34, 2014. [37] A. Bulat and G. Tzimiropoulos, "Hierarchical Binary CNNs for
[31] D. Dasgupta, S. Yu and F. Nino, “Recent advances in artificial Landmark Localization with Limited Resources," IEEE Transactions
immune systems: Models and applications,” Applied Soft Computing, on Pattern Analysis & Machine Intelligence, vol. 42, no. 02, pp. 343-
vol. 11, no. 2, pp. 1574-1587, 2011. DOI: 10.1016/ 356, 2020. DOI: 10.1109/TPAMI.2018.2866051.
j.asoc.2010.08.024. [38] V.A. Sindagi and V.M. Patel, “A survey of recent advances in CNN-
[32] Y. Li, X. Zhang and D. Chen, “CSRNet: Dilated convolutional neural based single image crowd counting and density estimation,” Pattern
networks for understanding the highly congested scenes,” Proc. IEEE Recognit. Lett., vol. 107, pp. 3-16, 2017.
Conf. Comput. Vis. Pattern Recognit, pp. 1091-1100, 2018. [39] K. He, G. Gkioxari, P. Dollar and R. Girshick, "Mask R-CNN," IEEE
[33] M. Kalayeh and M. Shah, "Training Faster by Separating Modes of Transactions on Pattern Analysis & Machine Intelligence, vol. 42, no.
Variation in Batch-Normalized Models," IEEE Transactions on 02, pp. 386-397, 2020. DOI: 10.1109/TPAMI.2018.2844175.
Pattern Analysis & Machine Intelligence, vol. 42, no. 6, pp. 1483- [40] S. Lin, R. Ji, C. Chen, D. Tao and J. Luo, "Holistic CNN
1500, 2020. DOI: 10.1109/TPAMI.2019.2895781R. Compression via Low-Rank Decomposition with Knowledge
[34] A. Farrugia and C. Guillemot, "Light Field Super-Resolution Using a Transfer," IEEE Transactions on Pattern Analysis & Machine
Low-Rank Prior and Deep Convolutional Neural Networks," IEEE Intelligence, vol. 41, no. 12, pp. 2889-2905, 2019. DOI:
Transactions on Pattern Analysis & Machine Intelligence, vol. 42, no. 10.1109/TPAMI.2018.2873305.
05, pp.1162-1175, 2020. DOI: 10.1109/TPAMI.2019.2893666. [41] I. Rocco, R. Arandjelovic and J. Sivic, "Convolutional Neural
[35] Z. Zhou, J. Shin, L. Zhang, S. Gurudu, M. Gotway and J. Liang, Network Architecture for Geometric Matching," IEEE Transactions
“Fine-tuning convolutional neural networks for biomedical image on Pattern Analysis & Machine Intelligence, vol. 41, no. 11, pp.
analysis: Actively and incrementally,” Proc. IEEE Conf. Comput. 2553-2567, 2019. DOI: 10.1109/TPAMI.2018.2865351.
Vis. Pattern Recognit., pp. 7340-7349, 2017.
VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 101