=Paper=
{{Paper
|id=Vol-2667/paper49
|storemode=property
|title=An integrated approach to mapping user profiles on social networks
|pdfUrl=https://ceur-ws.org/Vol-2667/paper49.pdf
|volume=Vol-2667
|authors=Vladimir Belov,Dmitriy Drozdov,Roman Shakurov,Vadim Moshkin,Ilya Andreev
}}
==An integrated approach to mapping user profiles on social networks ==
https://ceur-ws.org/Vol-2667/paper49.pdf
An integrated approach to mapping user profiles on
social networks
Vladimir Belov Dmitriy Drozdov Roman Shakurov
Information Systems department, Information Systems department, Information Systems department,
Faculty of Information Systems and Faculty of Information Systems and Faculty of Information Systems and
Technologies Technologies Technologies
Ulyanovsk State Technical University Ulyanovsk State Technical University Ulyanovsk State Technical University
Ulyanovsk, Russia Ulyanovsk, Russia Ulyanovsk, Russia
belo.199666@mail.ru dimoxx123@gmail.com relife@inbox.ru
Vadim Moshkin Ilya Andreev
Information Systems department, Faculty of Information Systems Information Systems department, Faculty of Information Systems
and Technologies and Technologies
Ulyanovsk State Technical University Ulyanovsk State Technical University
Ulyanovsk, Russia Ulyanovsk, Russia,
v.moshkin@ulstu.ru ia.andreev@ulstu.ru
Abstract—In this paper, we consider an integrated approach II. THE MAIN APPROACHES TO SOLVING THE PROBLEM OF
for the sole identification of a person in several different social USER IDENTIFICATION
networks by analyzing the questionnaire data, poorly structured
information and images comparison from the profiles of social A. Methods and algorithms for mapping user accounts on
networks. Also the paper contains the description of the social networks
software service that implements the proposed approach.
Currently, the task of identifying users using data profiles
Keywords—social network, account, search, mapping of social networks is solved in various ways [1, 2, 3].
In [4, 5, 6, 7, 8], methods for analyzing data profiles of the
I. INTRODUCTION social networks MySpace, StudiVZ are described. But these
The active growth of the audience of social networks has networks are not popular in Russia. The proposed approaches
led to the emergence of these resources as a new source of data consist in constructing feature vectors of user characteristics
and knowledge. In Russia, several social networks are based on the information provided on personal pages. To the
currently the most popular, each of which has its own focus obtained vectors, methods of exact, partial and fuzzy
and specificity of the content posted. Such resources include comparison are applied. In these works, the authors proposed
VKontakte, Odnoklassniki, Instagram, Facebook, Youtube, features that are most significant when comparing accounts.
Twitter[1]. Many users have several accounts on different The developed algorithms the accuracy of about 80% on a test
social networks and publish different or similar content to sample of user accounts.
them. And to find a person in any of the networks becomes
problematic. In [6, 7, 9], methods for mapping user profiles of social
networks by analyzing published unstructured (text)
Working with social networks can be beneficial in information are presented. In [6], the authors conclude that the
implementing the functions of the company’s personnel creator of a text note can be identified by a unique writing
management system, as you can often find out much more style. In [7], a method is shown that takes into account not
information about the professional and personal qualities of only text information published by a user in a note, but also
the applicant from social networks than from the CV. meta-information associated with it: geolocation, publication
Currently, the collection or meaningful analysis of time, hashtags, etc.
information from social networks is carried out manually by
specialists of personnel services, which is time consuming and B. Software services for searching users in social networks
limits the amount of information processed. Currently, there are several services for searching for
profiles of people in social networks in RuNet. Most services
Thus, there is a need to develop a software system that work on the principle of conventional search engines -
allows you to identify a person’s profile in several social download all available open profile data and save it to a local
networks. Such developments would allow aggregating more database.
data about users to assess the severity of their personal
characteristics. This work is aimed at solving the problem of FindFace [10]. One of these services is the FindFace
searching for an integrated approach for mapping (comparing) system, as well as many other systems based on it, which
user profiles in various social networks based on the analysis allow you to find a person’s profile on a social network from
of structured data, text information, as well as graphic their photo. To start the search, you need to select a photo
materials for the purpose of further analysis of the user's social where the human face is clearly visible, and upload the
portrait. picture. The algorithm will find pages with similar photos and
lay out links to them with examples of images. Each link will
have a rating from 0 to 1. If the indicator is more than 0.67,
then this means that the system recorded the most complete
match. The developed neural network scanned the faces of
500 million users of the VKontakte social network.
Copyright © 2020 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0)
Data Science
Yandex.People [11]. The Yandex.People system uses text max
data obtained from social network profiles to search. So the
following data is uploaded from the profiles of a person: where is the number of pairwise matching lemmas, 1,
Name of the user (or at least one of the parameters 2 - the number of lemmas in lines 1 and 2, respectively.
allowing to identify the person). If the value of this criterion is more than 0.85, then the
lines are considered similar.
Age of user.
Criteria for the presence of similar posts. Two metrics
Place of residence or user address. were used to compare text notes. The first is finding
the Levenshtein distance (editorial distance, editing
Place of study or completed education. distance)[12, 13] - the minimum number of operations
User’s place of work. to insert one character, delete one character and
replace one character with another, necessary to turn
Despite the availability of these services that solve specific one line into another. As a second method for finding
tasks of searching for users of social networks, there are similar posts, the shingles algorithm was
currently no comprehensive approaches and universal services implemented [14]. This algorithm works on the
that allow users to compare user profiles in various social principle of splitting text into shingles, computing
networks by analyzing not only the data of profiles, but also hashes of data of shingles, pairwise comparison of
poorly structured information from the pages of the respective hashes. The following metric was used for the shingle
accounts. method:
III. APPROACH TO MAPPING USER PROFILES ON SOCIAL
NETWORKS
The developed algorithm takes as a basis a person’s where is the number of matching hashs of shingles, ,
personal page from a social network. Different information is - the number of shingles in the first and second row,
downloaded and is used for further search and comparison of respectively.
the profile in various social networks. At the current stage, the
following data is used: A visual representation of the shingle algorithm is shown in
Figure 1.
First name, middle name of the user;
Date of Birth;
Place of residence;
Place of Birth;
Friends;
Text notes (posts);
Place of work;
Place of study;
Fig. 1. Shingles Algorithm.
Contacts, email, phone number; Criteria for having similar friends. This indicator is
Profile avatar, as well as profile photos. calculated by pairwise comparison of the names of
friends. The more matches, the higher the profile in
This information is downloaded, both for the original the final search results.
profile, and for the desired profiles in other social networks.
The loaded profile data is mapped to the source profile data. IV. IMPLEMENTATION OF A SOFTWARE SYSTEM FOR MAPPING
Since there can be several profiles found, they are sorted USER PROFILES ON SOCIAL NETWORKS
according to the following criteria:
To test the effectiveness of the proposed approach, a
Criteria for the presence of similar faces in software system for mapping user profiles on social networks
photographs. Using the HOG method, people are was implemented. The developed system is a client-server
found in photographs and their vector representation application, where the server is a Java web service developed
is generated. Subsequently, the Euclidean norms of using the Spring Boot software platform.
the vectors are compared. The system integrates with the three most popular social
The criterion for the presence of similar contacts. networks in the CIS: Vkontakte [15], Odnoklassniki [16] and
Profiles containing links to each other are very likely Facebook [17]. Data from the VKontakte social network is
to belong to the same person. downloaded through the integration with VK API. Data from
the Odnoklassniki and Facebook is downloaded by parsing the
The criterion for a similar place of work and place of desktop and mobile versions of the website of the respective
study. To calculate this indicator, the strings are pre- networks.
processed: they are cleared of punctuation, reduced to
lower case. After that, the lines are lemmatized, and A web service containing an application in python has also
using the obtained lemmas, the lines are compared been developed. This service, using the DLIB library [18],
according to the following metric: forms a vector representation for user photos.
VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 226
Data Science
As input, the system accepts a link to a profile in one of The page displays the selected profiles, and on the left
the social networks. From this profile, all possible data about there is an application menu.
a person is loaded, and according to these data a single model
of the desired profile is formed. Similar profiles are searched V. EXPERIMENT RESULTS
by searching for fields loaded from the original profile. Based A pre-prepared sample of 100 users with profiles in
on loaded similar profiles, a rating is formed, according to various social networks was used as an experimental base. All
which sorting subsequently occurs. these users had 204 accounts, since not all of them had
Similar profiles are sorted by the received rating and accounts in all networks at once. For each of these accounts,
displayed on a web-form. we tried to find similar ones using the developed service. As a
result of the experiments, the diagram shown in Figure 4 was
The system architecture is shown in Figure 2. compiled.
Fig. 4. Chart of the percentage of found profiles.
On the diagram you can see that the system coped best
with finding profiles on the VK social network, and worst of
all, Facebook. This is due to the convenience of extracting
data from relevant resources. VK API allows you to quickly
Fig. 2. System architecture. extract large amounts of data, which increases the quality of
As can be seen from the figure, the system consists of two recognition, while parsing other networks consumes many
server applications: resources, which forces to limit the amount of data retrieved.
Spring Boot App [19], The results for the profile comparison criteria were also
calculated, the result is shown in Figure 5.
Flask App [20]
There is also a React App client application. Spring Boot
App contains the basic logic of the system, as well as data
loaders from social networks:
VK Api Loader,
Facebook loader,
OK loader
The Flask App contains methods for recognizing faces in
photographs using the DLIB library. H2 Database is used to
store some non-confidential data. An example of such data is
the id of cities and countries from the VK API. Fig. 5. Profile comparison criteria diagram.
An example of the system interface is shown in Figure 3. On the diagram you can see that in all cases the system
managed to find at least one common friend. The results of the
coincidence of other criteria are much smaller. Twice worse,
the system managed to find common educational institutions
and common places of work. This is due to the fact that users
do not always indicate these characteristics on their pages.
Also, often the format of the specified data does not allow to
correctly compare them. Even less, the system coped with
finding common faces in photos. This is due to many factors,
such as, for example, the accuracy of the model itself, the
quality and number of photos uploaded. Only a third of the
experiments managed to find common posts on the pages.
This is due to the fact that users do not always fill pages with
the same posts. Cross-references to profiles were found least
Fig. 3. System appearance.
of all, as users provide such information less often.
VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 227
Data Science
CONCLUSION [4] Y. Gaewon, "Enhancing Entity Search with Social Network
Matching," EDBT/ICDT: Proceedings of the 14th International
Thus, within the framework of this work, an integrated Conference on Extending Database Technology, 2011.
approach was proposed to find user profiles in different social [5] M. Motoyama and G. Varghese, "I seek you: searching and matching
networks by analyzing not only the data of profiles, but also individuals in social networks," Proceedings of the eleventh
poorly structured information from the pages of the respective international workshop on Web information and data management,
2009.
accounts, as well as graphic information.
[6] E Raad, R. Chbeir and A. Dipanda, "User profile matching in social
As a result of the work done, a software system was networks," 13th International Conference on Network-Based
developed that performs the function of searching and Information Systems, IEEE, 2010.
mapping similar profiles on social networks. The application [7] J. Vosecky, D. Hong and V.Y. Shen, "User identification across
can be used as a personnel search platform. The proposed multiple social networks," 1 international conference on networked
digital technologies, IEEE, 2009.
methodology lays the foundation for further work on
[8] I. Veldman, “Matching Profiles from Social Network Sites,” Master's
conducting relevant experiments, developing new algorithms thesis, University of Twente, 2009.
for searching, comparing, analyzing, and building a portrait of [9] N. Yarushkina, A. Filippov, V. Moshkin, A. Namestnikov and G.
a user based on open data about they. Guskov, “The social portrait building of a social network user based on
semi-structured data analysis,” CEUR Workshop Proceedings, vol.
ACKNOWLEDGMENT 2475, pp. 119-129, 2019.
This work was supported by the Foundation for Assistance [10] FindFace [Online]. URL: https://findface.pro.
to the Development of Small Forms of Enterprises in the [11] Yandex.People [Online]. URL: https://yandex.ru/people.
Scientific and Technical Sphere within the framework of the [12] V. Chernenkyi and Yu. Gapanyuk, “Passenger identification technique
project "Development, technical implementation and testing based on installation data,” Engineering Journal: Science and
Innovation, vol. 3, no. 3, pp. 3, 2012.
of a prototype platform for the formation of a social portrait of
[13] D.V. Mikhailov, A.P. Kozlov and G.M. Emelyanov, “An approach
an applicant based on intelligent data retrieval in social based on analysis of n-grams on links of words to extract the
networks using the principles of knowledge engineering" of knowledge and relevant linguistic means on subject-oriented text sets,”
Agreement No. 60GS1CTS10-D5 / 56043 from 06.02.2020. Computer Optics, vol. 41, no. 3, pp. 461-471, 2017. DOI: 10.18287/
2412-6179-2017-41-3-461-471.
REFERENCES [14] A.Tsimbalov and O. Zolotarev, “The method of shingles,” Vestnik of
[1] I. Rytsarev, D. Kirsh and A. Kupriyanov, “Clustering of media content the russian new university. Series: Complex systems: models, analysis
from social networks using bigdata technology," Computer Optics, vol. and management, vol. 4, no. 4, pp. 72-79, 2016.
42, no. 5, pp. 921-927, 2018. DOI: 10.18287/2412-6179-2018-42-5- [15] Vkontakte [Online]. URL: https://vk.com, last accessed 2020/05/11.
921-927.
[16] Odnoklassniki [Online]. URL: https://ok.ru/.
[2] A. Filippov, V. Moshkin and N. Yarushkina, “Development of a
Software for the Semantic Analysis of Social Media Content,” Recent [17] Facebook [Online]. URL: https://www.facebook.com.
Research in Control Engineering and Decision Making. Studies in [18] DLIB library [Online]. URL: http://dlib.net/.
Systems, Decision and Control, vol 199, pp. 421-432, 2019. [19] Spring Boot App [Online]. URL: https://spring.io/projects/spring-boot.
[3] N. Yarushkina, A. Filippov, V. Moshkin, G. Guskov and A. Romanov, [20] Flask App [Online]. URL: https://flask.palletsprojects.com/en/1.1.x
“Intelligent Instrumentation for Opinion Mining in Social /quickstart/.
Media,” Proceedings of the II International Scientific and Practical
Conference Fuzzy Technologies in the Industry, Ulyanovsk, Russia,
pp. 50-55, 2018.
VI International Conference on "Information Technology and Nanotechnology" (ITNT-2020) 228