Social Recommender Systems


Recommender systems play an increasingly important role in the success of social media websites. Higher portions of social websites’ traffic are triggered by recommendations and those sites rely on the quality of the recommendations to attract new users and retain existing ones. In this chapter, we will introduce the notion of social recommender systems as recommender systems that target the social media domain. After a short introduction, we will discuss in detail two of the most prominent types of social recommender systems — recommendation of social media content and recommendation of people. We will describe the main approaches and state-of-the-art techniques for each of the recommendation types. We will also review related work from the recent years that studied such recommender systems, in order to demonstrate the different use cases and methods applied to take advantage of the unique data. We will conclude by summarizing the key aspects, emerging domains, and open challenges for social recommender systems.


The recent decade introduced the “social web” or “Web 2.0” (o’Reilly 2009), a web where people play a central role by creating content, annotating it with tags, votes (or ‘likes’), or comments, joining communities, and connecting to friends. Social media websites are proliferating and attract millions of users who author content, post messages, share photos with their friends, and engage in many other types of activities. This rapid growth intensifies the phenomenon of social overload, where users of social media are exposed to a huge amount of information and participate in huge amounts of interactions. Social overload makes it harder on the one hand for social media users to choose which sites to engage in and for how long and on the other hand makes it more challenging for social media websites to attract users and retain them.

Social Recommender Systems (SRS) are recommender systems that target the social media domain. They aim at coping with the social overload challenge by presenting the most relevant and attractive data to the user, typically by applying personalization techniques. The “marriage” between recommender systems (RS) and social media has many potential benefits for both sides. On the one hand, social media introduces many new types of data and meta-data, such as tags and explicit online relationships, which can be used in a unique manner by RS to enhance their effectiveness. On the other hand, recommender systems are crucial for social media websites to enhance the adoption and engagement by their users and thus play an important role in the overall success of social media. It should be noted that traditional RS, such as user-based collaborative filtering, are social in their nature since they mimic the natural process where we seek advice or suggestions from other people (Resnick 1997). Yet, in this chapter we focus on those recommender systems that are aimed for the social media domain, which we term social recommender systems (Guy 2011).

This chapter focuses on two key areas of SRS, social media content recommendation and people recommendation. We dedicate a section to each of these areas, reviewing the different sub-domains, their unique characteristics, the applied methods, case studies in the enterprise, and open challenges. SRS consist of more areas, such as recommendation of tags and groups (communities), however, these are left beyond the scope of this chapter. The remainder of the chapter is organized as follows. The next two sections discuss in detail content and people recommendation. The following section discusses key aspects characterizing SRS as raised throughout its preceding two sections. The chapter concludes by reviewing emerging SRS domains and open challenges.

Content Recommendation

\label{sec:content} Social media introduced many new types of content that can be created and shared by any user in a way that has never been possible before. Users became the center of every social media website and in many cases were the ones creating the actual content of the site: textual content as in Wikipedia and WordPress; photos as in Flickr and Facebook; and video as in YouTube. Users also have a key role in providing feedback and annotating exiting content on social media websites. Comments allow users to add their own opinion; Votes and ratings allow them to ‘like’ (or dislike) favourite posts; and tags allow them to annotate the content with keywords that reflect their own viewpoint. These new types of feedback forms allow RS to implicitly infer user preferences and content popularity by analyzing the crowd’s feedback.

In the social media era, articulated relationships have become available through social network sites (SNSs) (Boyd 2007) and changed the world of content recommendation. While in the past such relationships could only be partially extracted by surveys and interviews, and later by mining communication patterns from phone logs or email that are highly sensitive privacy-wise, the availability of relationships in social networks allows tapping into one’s network of familiar people (Facebook, LinkedIn) or people of interest (Twitter) in a simpler way without infringing privacy. The use of the friend list instead of or alongside the list of similar people as in traditional CF has been broadly proven to be productive for enhancing content recommendations. Sinha and Swearingen (Sinha 2001) were among the first to compare friend-based recommendation with traditional methods and showed their effectiveness for movie and book recommendation. Golbeck (Golbeck 2006) showed that friends can be a trusted source for movie recommendations. Groh and Ehmig (Groh 2007) compared collaborative filtering with friend-based “social filtering” and showed the advantage of the latter for club recommendation within a German SNS. Overall, recommendation based on friends enhance recommendations’ accuracy; allow the user to better judge the recommendations since s/he is familiar with the respective people; spare the need for explicit feedback from the user in order to calculate similarity; and can be used for coping with the cold-start problem for new users.