SocConnect: A personalized social network aggregator and recommender

https://doi.org/10.1016/j.ipm.2012.07.006Get rights and content

Abstract

Users of Social Networking Sites (SNSs) like Facebook, LinkedIn or Twitter, are facing two problems: (1) it is difficult for them to keep track of their social friendships and friends’ social activities scattered across different SNSs; and (2) they are often overwhelmed by the huge amount of social data (friends’ updates and other activities). To address these two problems, we propose a user-centric system called “SocConnect” (Social Connect) for aggregating social data from different SNSs and allowing users to create personalized social and semantic contexts for their social data. Users can blend and group friends on different SNSs, and rate the friends and their activities as favourite, neutral or disliked. SocConnect then provides personalized recommendation of friends’ activities that may be interesting to each user, using machine learning techniques. A prototype is also implemented to demonstrate these functionalities of SocConnect. Evaluation on real users confirms that users generally like the proposed functionalities of our system, and machine learning can be effectively applied to provide personalized recommendation of friends’ activities and help users deal with cognitive overload.

Highlights

► SocConnect allows users to define personal contexts of social data from SNSs. ► It also allows users to indicate their interest levels for social activities. ► SocConnect provides personalized recommendation of social activities to users. ► We suggest a ML method and a set of features for learning user preferences.

Introduction

The advent of web 2.0 technology especially Social Networking Sites (SNSs), has changed the way people communicate. Clara Shih, in her book “The Facebook Era” (Shih, 2009), observes that social media such as Facebook (facebook.com) have transformed the socio-cultural landscape – people’s behaviour, attitudes, interactions, and relationships. People spend more time on SNSs than ever, and prefer communication via SNSs over emails (Chisari, 2009). Every successful SNS has its unique features. Facebook allows a large number of third party applications to build on its APIs. Twitter (twitter.com) offers micro-blogging and an asymmetric following relation between users. MySpace (myspace.com) has a large user community interested in music. LinkedIn (linkedin.com) focuses on career and professional networking. Despite the diversity of SNSs and the fact that social media enriches people’s lives, current SNSs have several significant limitations (Erétéo, Buffa, Gandon, Leitzelman, & Limpens, 2009), two of which motivate our work.

In the context of SNS, the “walled garden” problem is about the SNSs companies such as Facebook or Twitter having control over user’s data. With the explosion of the number of SNSs, it is also common that one user engages with multiple SNSs. In July 2009, Anderson Analytics conducted an online survey1 over 11,000 SNS users. The results show a high overlap of user populations of Facebook, Twitter and LinkedIn. User-generated content, users’ online activities, and their friendships are scattered over different SNSs. It becomes increasingly inconvenient for users to manage their social data and constantly check several SNSs to keep track of all recent updates. Even worse, people may have different accounts on the same SNS.

Another problem of SNSs is information overload. The users of multiple SNSs see a large amount of social data generated by their network friends everyday. In this work, “social data” denotes status updates, posts of photos, links, likes, retweets, i.e. all new items that appear in the stream of updates in a SNS. The innovation of SNS has constantly increased the richness of social data. This causes significant information overload to users. Christian Kreutz in his blog described this specified kind of information overload as “network overload”.2 Network overload is caused by two reasons: first, there is too much new social data appearing constantly on SNSs; second, social data often does not have explicit context. The first reason is fairly intuitive, but the second one needs some explanations.

SNSs generate huge amount of social data. However, lots of the data do not have explicit context. For example, the way the word “friend” is used in Facebook does not reflect the true meaning of the word in colloquial English. On Facebook, a user’s “friends” may include co-workers, college mates, and people whom the user barely knows but was too polite to decline their invitation. It is thus important to have a way of distinguishing these people. Users and their friends on different social networking sites may also have different kinds of relationships. For example, Facebook friends are mostly people whom the user already knows (Lampe, Ellison, & Steinfield, 2006), but users may have not met most of their Twitter friends in person. Without explicit context, it becomes very difficult to handle the huge amount of social data and too complex for users to make sense of the data. The contexts may include the type of social bound (the provenance, closeness, symmetry, etc.), the type of relationships (family, colleagues and friends in personal life), the common interests they share, the closeness of friendships, and the location of friends.

The “network overload” problem becomes more serious when the social data of the user is aggregated across different SNSs into one place by a social aggregator application. A social network aggregator is the application pulls together content from multiple social network service into a single location. The number of updates will increase significantly in this case. One way to deal with information overload is by providing recommendations for interesting social updates, which allows the user to focus her attention more effectively.

In this paper, we propose a system called “SocConnect” (short for social connect) which attempts to address these two problems, “walled garden” and “network overload”. SocConnect provides functionality to integrate social data across SNSs, and to allow users to organize their aggregated social data. The users can define social contexts of their social data. The added context can then help users to browse their social data. Moreover, SocConnect learns the users’ preferences using machine learning techniques and recommends new unread social data to them based on their preferences. As the evaluation of the effectiveness of our system, we collect data from real users to show the good performance on personalized recommendations of social data, and that users generally like the proposed functionalities of our system.

The rest of the paper is organized as follows. Section 2 provides a summary of related work on social network aggregators and recommendation. Section 3 presents the proposed schema used by SocConnect to integrate social data across SNSs, and the main functionalities of SocConnect. Section 5 describes a prototype of SocConnect to demonstrate its functionalities. Section 6 evaluates the effectiveness of the personalized recommendation functionality and the usability of all the functionalities. Finally, Section 7 summarizes the contributions of our current work and proposes some future directions.

Section snippets

Related work

SocConnect is a social network aggregator and recommender. Here, we discuss important requirements for integrating social data across different SNSs and compare with other existing social network aggregators. We also survey the state-of-art recommender systems and clearly point out that our recommendation is content-based and shares similarity with text recommendation.

SocConnect

In this section, we first present a schema for representing integrated social data across SNSs. We then describe in details the proposed functionalities of our SocConnect system and the implementation of the functionalities.

Personalized recommendations in SocConnect

Our approach of personalized recommendation in SocConnect is content-based rather than collaborative. In this section, we propose a list of potential non-textual and textual features for representing each activity and present several machine learning techniques used to predict users’ preferences on activities from the social networking sites of Twitter and Facebook.

Demonstration of SocConnect

We provide several screenshots to demonstrate the user interface of SocConnect. This interface is an early prototype implementing the main functionalities rather than the ultimate interface for SocConnect. We use Facebook and Twitter for the purpose of demonstration. Suppose that a user Jane has accounts on both Facebook and Twitter. SocConnect retrieves Jane’s social data on these two sites. The social data of her friends can then be managed, browsed and filtered by her SocConnect dashboard

Performance of personalized recommendation

We first carried out experiments to evaluate (1) the performance of the four machine learning techniques for learning user preferences on social activities and (2) the performance of personalized recommendations when different features are used to represent social activities. Social data streams from ten subjects were used in the evaluation. Five of the subjects are from Saskatoon, Canada, and the other five are from New Jersey, USA. Half of them are students and the other half are workers. Six

Contribution and future work

In this work, we proposed the SocConnect system to personalized aggregation and recommendation of social data from different social networking sites, to address the two important problems faced by SNS users, “walled garden” and “network overload”. SocConnect provides a set of functionalities, including blending and grouping friends, tagging friends and social activities, and the personalized recommendations of social activities. Results of our user study indicate strong support for the

References (17)

  • C. Shih

    The Facebook Era: Tapping online social networks to build better products, reach new audiences, and sell more stuff

    (2009)
  • Chisari, M. (2009). The future of social networking. In Proceedings of the W3C workshop on the future of social...
  • Erétéo, G., Buffa, M., Gandon, F., Leitzelman, M., & Limpens, F. (2009). Leveraging social data with semantics. In...
  • Lampe, C., Ellison, N., & Steinfield, C. (2006). A face(book) in the crowd: Social searching vs. social browsing. In...
  • Heckmann, D., Schwarzkopf, E., Mori, J., Dengler, D., & Kroner, A. (2007). The user model and context ontology gumo...
  • Bojars, U., Passant, A., Breslin, J., & Decker, S. (2008). Social network and data portability using semantic web...
  • Chen, J., Nairn, R., Nelson, L., Bernstein, M., & Chi, E. (2010). Short and tweet: Experiments on recommending content...
  • Resnick, P., Lacovou, N., Suchak, M., Bergstrom, P., Riedl, J. (1994). Grouplens: An open architecture for...
There are more references available in the full text version of this article.

Cited by (28)

  • FLOWER: Fusing global and local associations towards personalized social recommendation

    2018, Future Generation Computer Systems
    Citation Excerpt :

    Consequently, such online social networks as Facebook, Twitter, Last.fm, Epinions, and YouTube, not only provide novel ways for people to communicate and share their resources and opinions with each other, but also serve as a platform to establish new relationships among people. Furthermore, to promote the evolution of social networks and meet the demand of interactions and communications, such online social networks as Google+, Facebook and Twitter, have provided the functionality of suggesting friends to people through utilizing their personalized information [24,25], making friend recommendation an increasingly important component in studies of social networks [26]. The first category includes methods considering existing relations simply to calculate ranking scores of candidates.

  • Why do people (not) like me?: Mining opinion influencing factors from reviews

    2017, Expert Systems with Applications
    Citation Excerpt :

    Many published studies propose healthcare-oriented recommendations. For instance, in Zhang, Wang, and Vassileva (2013), a content-based personalized recommendation system called SocConnect is proposed, and a collaboration-based medical knowledge recommendation system for clinicians is introduced by Huang, Lu, Duan, and Zhao (2012). For further information on recommendation systems in healthcare, see, Sanchez-Bocanegra, Sanchez-Laguna, and Sevillano (2015), Wiesner and Pfeifer (2014), and Zhang, Chen, Huang, Wu, and Li (2016).

View all citing articles on Scopus
View full text