Encyclopedia of Social Network Analysis and Mining

Living Edition
| Editors: Reda Alhajj, Jon Rokne

Time-Aware Egocentric Network-Based User Profiling

  • Sirinya On-AtEmail author
  • Marie-Françoise Canut
  • André Péninou
  • Florence Sèdes
Living reference work entry
DOI: https://doi.org/10.1007/978-1-4614-7163-9_110149-1

Keywords

Recommender System User Profile Online Social Network Collaborative Filter User Interest 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Synonyms

Glossary

Egocentric network

A special type of social network that involves a focal user called “ego.” The egocentric network of a user (ego) consists of the individuals called “alters” having a direct relationship with the user (ego) together with the relationships between these individuals

User profile

A record of user information (personnel data, preferences, interests). In information systems, adaptive information mechanisms (e.g., personalization, information access, recommendation) rely on user profiles to propose relevant content according to the user-specific needs

Social profile

A particular user profile in which the interests are extracted from the information of user social network members

Definition

Egocentric network-based user profiling consists in extracting the user interests from information shared by the members of his/her egocentric networks. The user’s profile built within this approach will be called “social profile.” Time-aware egocentric network-based user profiling consists in applying a time-aware method into the social profile building process in order to extract relevant and up-to-date user interests.

Introduction

It has been shown from several works that user’s social neighbors can be a meaningful source to infer his/her interests. Egocentric network-based user profiling consists in extracting user interests from the information of his/her social neighbors and building his/her so-called social profile.

However, it is still a challenge to obtain relevant and up-to-date user interests from user’s egocentric network, particularly in online social networks (OSNs) context. In OSNs, user online behavior quickly evolves over time. The information and relationships in an OSN can rapidly become obsolete for the user. Extracting the interests from user’s egocentric network without taking into account its evolving characteristic may lead to irrelevant long-term interests for the user.

As relationships and information in OSNs evolve over time, the main issues are: (i) how to select the relevant individuals in the user egocentric network as meaningful information sources? and (ii) how to select relevant information from the selected individuals to extract relevant interests?

This entry presents the related works of a time-aware egocentric network-based user profiling approach dealing with the evolution of user interests, related to the social network evolution issue.

Key Points

  • Definition of egocentric network-based user profiling

  • Different approaches of egocentric network-based user profiling

  • Incorporating interests dynamic characteristic in the profile

  • Taking into account OSN evolution in the egocentric network-based social profiling process

Historical Background

Egocentric Network-Based User Profiling

With social content sharing explosion, the amount of available online information becomes a rich source of knowledge nevertheless sometimes superfluous for a user. User profiling becomes essential for adaptive systems (e.g., personalization system, recommendation system) to identify user information in order to propose relevant content according to his/her specific needs. Different user profiling approaches are discussed in (Gauch et al. 2007).

Recently, different works have shown that user’s social neighbors can be a meaningful source to infer his/her interests. Besides, sociology works (Raghavan 2002; Sinha and Swearingen 2002) have shown that the user is better described by people around him/her, especially the people that are directly connected to him/her (his egocentric network). The term “egocentric network-based user profiling” is considered to refer to the interest extraction approach that consists in extracting user interests from information of his/her social neighbors. The user’s profile built within this approach will be called “social profile.” Social profile can be used as an additional profile to the classical user profile for new or less active users whose profile are empty or poor for the system.

Several works have been proposed to prove and improve the relevance of the egocentric network-based user profiling approach. These works can be distinguished in two approaches:
  • Individual-based approach that extracts user interests and computes their score according to the characteristic of each individual. Zeng et al. (2009) proposed a social personalized information retrieval system of scientific papers on DBLP. They improved the relevance of personalized search results for each author by taking into account the interests extracted from his/her coauthors. Wen and Lin (2010, 2011) proposed an approach that combines topic similarity among users’ online social networks content with network features such as familiarity with their social neighbors in order to improve the quality of inferring their interests. Carmel et al. (2009) improved search engine queries results by taking into account the user directed social neighbors to calculate the relevant score of each document (result of queries). Cabanac (2011) proposed a social recommender system in a coauthors network that uses a coauthors graph and a graph of venue (in conferences) to recommend relevant coauthors to a researcher. In this work, people are selected individually according to their topical similarity, their proximity, their connectivity, or the strength of their tie in their social graph.

  • Community-based approach that extracts user interests from extracted communities and computes their score according to the characteristics of each community. Mislove et al. (2010) tried to uncover a user’s hidden attribute from partial friends’ opened profiles on Facebook. They applied a community detection algorithm and inferred user attributes based on similarity to other members of the same community. Tchuente et al. (2013) proposed to extract the communities from the user’s egocentric network and generated user interests according to the characteristics of each community. The effectiveness of this work has been proved on coauthorship networks compared to the individual-based approach.

User Interests Evolution

As user behaviors evolve over time, it is necessary to take into consideration the evolution of user interests in user profiling process. In fact, the user profile at a moment should take into account the changing behavior of each user interest from any previous moment (i.e., determine if an interest becomes more or less meaningful for the user at the moment).

The temporal factor is so viable to incorporate this issue. Different works in user interests mining have been proposed in this context. Some works rely on the time window approach that selects only the information from the latest periods (Maloof and Michalski 2000). In this approach, outdated information is completely forgotten and can lead to the loss of useful knowledge when time shift is gradual. To avoid this drawback, some works propose a time decay approach. This approach consists in weighting different time periods according to their relevance. To perform this computation, time decay functions which assign the higher weights to the most recent information are widely used (Li et al. 2013, 2014). This technique enables the use of all available information in a restricted way. In fact, the relevance of information can be computed by weighting their importance using some temporal factors. Another technique relies on ensemble learning to generate a family of predictors. Each predictor is weighted by its relevance according to the present time point (e.g., predictors that were more successful on recent instances get higher weights). Koren (2009) proposed to model the evolution of user behavior during the whole time period for collaborative filtering and demonstrates the effectiveness of his contribution for movies recommendation.

In the case of social profile building process, user interests are extracted from the information shared by his/her social neighbors. Hence, the evolution of extracted interests is related to the evolution of information shared on user social network and to the evolution of relationships between the user (ego) and his/her social neighbors (alters). This issue becomes particularly important in the OSNs context where user behavior evolves quickly. For a user, the relationships and information in his/her social network can evolve and become obsolete for him/her overtime.

Two users creating a relationship are not required to know each other in real life. As a result, the relationship persistence is not always maintained in this case. Furthermore, social events or viral marketing (buzz) are also factors that enhance online social content sharing. Often, these social phenomena occur for a short time period, then disappear and may reappear in another period.

Time-Aware Egocentric Network-Based User Profiling

As previously mentioned, user interests evolution has been taken into consideration in several works on user interests extraction processes. However, most of these works propose to take into account temporal factors to extract the dynamic interests from users’ own information (classical user profiling). As far as we know, very few works have been done to take into account the social network evolution in the egocentric network-based user profiling context.

Based on our literature review, the closest approach can be found in a work on a time-aware user-based collaborative filtering (CF). Yuan et al. (2013) proposed to apply a temporal factor to Point-of-Interests (POI) recommendation, which consists in recommending places that users have not visited before. This work extends the user-based CF model by applying a temporal influence and a geographical influence. In term of temporal influence, they first calculated the similarity between two users by using the check-in timestamp. Then, to recommend the places for a user, they considered the historical check-ins of his/her similar users at the moment of recommendation, rather than their check-ins record at all time.

Another aspect consists in applying a dynamic community extraction process to detect the trends in user generated contents (Cazabet et al. 2012). The authors defined an evolving network of terms by extracting different terms from user contents and created links between them. Each time a new term appears, it will become a new node. The authors applied a dynamic community detection algorithm, which takes as an input, a dynamic network and gave as a result a set of dynamic communities of terms. The terms from the extracted communities are then considered as the detected trends. Note that this technique is applied on the whole data on the network and does not fit exactly to the context of egocentric network-based user profiling. In fact, the sharing information can come from any users on the whole network who may not have common interests with the focal user. However, this technique may be envisaged to be applied on the information shared by the user social neighbors to obtain more restricted and relevant interests to build the user social profile.

The time-aware techniques adopted in the classical user profiling processes can be also envisaged on social profile building processes (i.e., instead of applying the techniques to users’ own information, we can apply them on the information shared by their neighbors). Though, it is necessary to adapt the techniques to the context of egocentric network-based approach. It means that we have to take into account the fact that the volume of information shared from social neighbors is much more important than his/her own information, and this information can evolve and become obsolete for the user overtime. Furthermore, each neighbor does not have the same capacity to influence the user, and their influence capacity can also change as time goes by.

In order to incorporate interests evolution in egocentric network profiling process and to take into account both relationships and information dynamics, Canut et al. (2015) proposed a time-aware user egocentric network-based user profile building method using the community-based approach. This work adopts the instance weighting technique which allows to use all available information in a differentiated manner by applying temporal factors. Considering the importance of the individuals of user social networks and their information, they proposed to integrate temporal criteria to weight both individuals and information in order to extract relevant and up-to-date user interests. A temporal score of an interest, represented by an element e extracted from an information Info indiv shared by an individual Indiv, is computed by using, on the one hand, the temporal relevance of Info indiv (Information temporal score) and, on the other hand, the temporal relevance of the individual Indiv (Individual temporal score). The information temporal score is computed regarding the freshness of Info indiv and the freshness of the last interaction between Indiv and the central user u. To reach this goal, a time exponential function f(t) = e λt proposed in (Ding et al. 2009) is adopted. This function assigns the higher weights to the most recent information. The value of t represents the elapsed time between the information timestamp and a given timestamp. λ∈ [0,1] represents the time decay rate. The higher λ is, the less important the old information is.

The individual temporal score is computed regarding the relationship strength between Indiv and the central user u. To achieve this, a time-aware link prediction metric based on the work of Tylenda et al. (2009) is adopted to compute a similarity score between each given Indiv and u. This score will represent their relationship persistence and approximate their relationship strength. The time-aware link prediction technique proposes the integration of the time exponential function into the existing Adamic/Adar metric (Liben-Nowell and Kleinberg 2003).

Finally, the final temporal score of an element e extracted from the information Info shared by an individual Indiv is computed by combining the information temporal score and the individual temporal score, varying with a parameter γ: γ IndivTempScore(e) + (1 − γ) InfoTempScore (e). Once the temporal score of all elements is computed, the elements are aggregated using the combination method CombMNZ, proposed by Shaw et al. (1994), to combine different weights from different communities. Finally, the aggregated element can be derived to the social profile as weighed user interests.

The experimentations on the coauthorship network (DBLP) have shown the effectiveness of this method compared to the existing time-agnostic one. This demonstrates the benefit of taking into account the evolution of user interests (by considering the social network evolution) in the social network-based user profiling process. The authors observed that in the coauthorship network context, the time decay rate of user relationship is required to be higher than the one of the shared information. They also observed that the individual temporal score has a larger preference over the information temporal score in the final temporal score calculation.

On-At et al. (2016) proposed to apply on Twitter the existing time-aware social profiling method proposed in Canut et al. (2015). With the particular social network characteristics of Twitter, the obtained results are different than those obtained on coauthorship networks. The community-based social profiling approach, which relies on the relationships between egocentric members, does not perform well in this kind of social networks. For the individual-based social profiling approach, the time-aware method can slightly improve the time-agnostic profiling method. The authors found that the relationship dynamics has weak impact over the social profiling process on Twitter compared to the information dynamics. Furthermore, they observed that the proposed method gives relevant results in terms of precision and recall on very sparse networks or rather dense networks. This observation needs more studies to find out better explanations for these results.

Key Applications

User profile is widely used in information adaptive mechanisms (e.g., recommendation system, personalization system) to collect and identify user information and interests in order to propose relevant content according to user-specific needs. For new or inactive users, their profile can be empty or does not contain any useful interests for a mechanism of personalization or recommendation. This problem is identified as the cold start problem (Massa and Avesani 2007). In such a case, the egocentric network-based user profiling can be applied to provide a social profile as an additional profile to complete the missing information.

Time-aware egocentric network-based user profile can provide potentially useful enhancement for online applications in which shared information and/or user relationships evolve quickly over time. For example, a user u is not football fan but supports occasionally his national team during the world cup, the well-known sport event that takes place every 4 years. He follows the official account of his favorite team and/or players on Twitter, to follow updated information about this team and the players. After the world cup, the information shared from these accounts become less meaningful for him. If this fact is not taken into account, the adaptive mechanisms that exploit his profile (e.g., personalized news recommendation) will continue to propose him the content concerning football, world cup, while this information becomes actually useless to him. For such scenarios, time-aware approach becomes essential to build the more relevant and up-to-date social profile.

Future Directions

The existing works have proved that the time-aware method can improve the relevance of user social profile compared to the time-agnostic method. To go a step forward and have better results, it is essential to improve the algorithms and/or temporal score calculation techniques. To reach this goal, other time-weight functions or link prediction algorithms can be applied. In addition, it has been shown that the time-aware egocentric network-based user profiling method performs differently regarding the characteristics of egocentric network (e.g., Twitter, dblp). Therefore, the remaining challenge consists in finding out an effective user interests extraction for each type of social network.

Cross-References

References

  1. Cabanac, Guillaume (2011) Accuracy of Inter-Researcher Similarity Measures Based on Topical and Social Clues. Scientometrics 87(3):597–620. doi: 10.1007/s11192-011-0358-1 CrossRefGoogle Scholar
  2. Canut Marie-Françoise, Sirinya On-At, André Péninou, Florence Sèdes (2015) Time-aware egocentric network-based user profiling. In: Proceedings of the 2015 IEEE/ACM international conference on advances in social networks analysis and mining 2015, ASONAM ’15. ACM, New York, pp 569–572. doi: 10.1145/2808797.2809415
  3. Carmel, David, Naama Zwerdling, Ido Guy, Shila Ofek-Koifman, Nadav Har’el, Inbal Ronen, Erel Uziel, Sivan Yogev, and Sergey Chernov (2009) Personalized Social Search Based on the User’s Social Network. In: Proceedings of the 18th ACM Conference on Information and Knowledge Management, 1227–1236. CIKM ’09. New York, NY, USA: ACM. doi: 10.1145/1645953.1646109
  4. Cazabet R, Takeda H, Hamasaki M, Amblard F (2012) Using dynamic community detection to identify trends in user-generated content. Soc Net Anal Mining 2(4):361–371. doi: 10.1007/s13278-012-0074-8 CrossRefGoogle Scholar
  5. Ding Y, Jacob EK, Caverlee J, Fried M, Zhang Z (2009) Profiling social networks: a social tagging perspective. D-Lib Mag 15(3/4). doi: 10.1045/march2009-ding
  6. Gauch S, Speretta M, Chandramouli A, Micarelli A (2007) User profiles for personalized information access. In: Brusilovsky P, Kobsa A, Nejdl W (eds) The adaptive Web, Lecture notes in computer science 4321. Springer, Berlin/Heidelberg, pp 54–89. http://link.springer.com/chapter/10.1007/978-3-540-72079-9_2 CrossRefGoogle Scholar
  7. Koren Yehuda (2009) Collaborative filtering with temporal dynamics. In: Proceedings of the 15th ACM SIGKDD international conference on knowledge discovery and data mining, KDD ’09. ACM, New York, pp 447–456. doi: 10.1145/1557019.1557072
  8. Li D, Cao P, Guo Y, Lei M (2013) Time weight update model based on the memory principle in collaborative filtering. J Comp 8(11):2763–2767. doi: 10.4304/jcp.8.11.2763-2767 Google Scholar
  9. Li L, Zheng L, Yang F, Li T (2014) Modeling and broadening temporal user interest in personalized news recommendation. Exp Sys App 41(7):3168–3177. doi: 10.1016/j.eswa.2013.11.020 CrossRefGoogle Scholar
  10. Liben-Nowell David Jon Kleinberg(2003) The link prediction problem for social networks. In: Proceedings of the twelfth international conference on information and knowledge management, CIKM ’03. ACM, New York, pp 556–559. doi: 10.1145/956863.956972
  11. Maloof MA, Michalski RS (2000) Selecting examples for partial memory learning. Mach Learn 41(1):27–52. doi: 10.1023/A:1007661119649 CrossRefzbMATHGoogle Scholar
  12. Massa Paolo, Paolo Avesani (2007) Trust-aware recommender systems. In: Proceedings of the 2007 ACM conference on recommender systems, RecSys ’07. ACM, New York, pp 17–24. doi: 10.1145/1297231.1297235.
  13. Mislove Alan, Bimal Viswanath, Krishna P. Gummadi, Peter Druschel (2010) You are who you know: inferring user profiles in online social networks. In: Proceedings of the third ACM international conference on Web search and data mining, WSDM ’10. ACM, New York, pp 251–260. doi: 10.1145/1718487.1718519
  14. On-At, S., A. Quirin, A. Péninou, N. Baptiste-Jessel, M. F. Canut, and F. Sèdes (2016) Taking into Account the Evolution of Users Social Profile: Experiments on Twitter and Some Learned Lessons. In: 2016 IEEE Tenth International Conference on Research Challenges in Information Science (RCIS), 1–12Google Scholar
  15. Raghavan P (2002) Social networks: from the Web to the enterprise. IEEE Internet Comp 6(1):91–94. doi: 10.1109/4236.989007 CrossRefGoogle Scholar
  16. Shaw Joseph A, Edward A Fox, Joseph A Shaw, Edward A Fox (1994) Combination of multiple searches. In: The Second Text REtrieval Conference (TREC-2), pp 243–252Google Scholar
  17. Sinha Rashmi R, Kirsten Swearingen (2002) Comparing recommendations made by online systems and friends. In: DELOS workshop: personalisation and recommender systems in digital libraries. http://dblp.uni-trier.de/db/conf/delos/delos2001.html#SinhaS01
  18. Tchuente D, Canut M-F, Jessel N, Peninou A, Sèdes F (2013) A community-based algorithm for deriving users’ profiles from egocentrics networks: experiment on Facebook and DBLP. Soc Net Anal Mining 3(3):667–683. doi: 10.1007/s13278-013-0113-0 CrossRefGoogle Scholar
  19. Tylenda Tomasz, Ralitsa Angelova, Srikanta Bedathur (2009) Towards time-aware link prediction in evolving social networks. In: Proceedings of the 3rd workshop on social network mining and analysis, SNA-KDD ’09. ACM, New York, pp 9:1–9:10. doi: 10.1145/1731011.1731020
  20. Wen Zhen, Ching-Yung Lin (2010) On the quality of inferring interests from social neighbors. In: Proceedings of the 16th ACM SIGKDD international conference on knowledge discovery and data mining, KDD ’10. ACM, New York, pp 373–382. doi: 10.1145/1835804.1835853
  21. Wen Zhen, Ching-Yung Lin (2011) Improving user interest inference from social neighbors. In: Proceedings of the 20th ACM international conference on information and knowledge management, CIKM ’11. ACM, New York, pp 1001–1006. doi: 10.1145/2063576.2063720
  22. Yuan Quan, Gao Cong, Zongyang Ma, Aixin Sun, Nadia Magnenat-Thalmann (2013) Time-aware point-of-interest recommendation. In: Proceedings of the 36th international ACM SIGIR conference on research and development in information retrieval, SIGIR ’13. ACM, New York, pp 363–372. doi: 10.1145/2484028.2484030
  23. Zeng Y, Y Yao, N Zhong (2009) DBLP-SSE: a DBLP search support engine. In: IEEE/WIC/ACM international joint conferences on Web intelligence and intelligent agent technologies, 2009, WI-IAT ’09, 1: pp 626–630. doi: 10.1109/WI-IAT.2009.364

Copyright information

© Springer Science+Business Media LLC 2017

Authors and Affiliations

  • Sirinya On-At
    • 1
    • 2
    Email author
  • Marie-Françoise Canut
    • 1
    • 2
  • André Péninou
    • 1
    • 2
  • Florence Sèdes
    • 1
    • 2
  1. 1.Toulouse Institute of Computer Science ResearchUniversity of Toulouse, CNRS, INPT, UPS, UT1, UT2JToulouseFrance
  2. 2.Systèmes d’Informations GénéralisésInstitut de Recherche en Informatique de Toulouse (IRIT)ToulouseFrance

Section editors and affiliations

  • Tansel Ozyer
    • 1
  • Ozgur Ulusoy
    • 2
  1. 1.TOBB Economics and Technology UniversityAnkaraTurkey
  2. 2.Bilkent UniversityAnkaraTurkey