Personalized recommendation in social network sites.

Abstract

Social network sites have experienced an explosion in both the number of users and the amount of user-contributed content in recent years. Today, millions of people visit Facebook, Twitter and the like to keep up with friends, to engage in random chatter, and to share and consume photos, news, useful tips and fun stories. Many active users of social network sites are, however, constantly troubled by information overload – there are too many other people to interact with and too much content to read. As a result, the difficulty in finding the right people and content to focus on has been identified as a key challenge for social network sites. In this thesis we seek to meet the challenge by designing three personalized recommender systems. The first system is a people recommender, which helps users find potential friends to connect with on social network sites; the second system is an information recommender, which helps users find interesting pieces of information from their online social network; the third system is a conversation recommender, which helps users find interesting conversations happening around their online social network. In designing the recommender systems, we adapt algorithms from related domains and create new algorithms. We refine these algorithms through offline data analysis. We then design and improve user interfaces of the recommender systems through participatory design. With the full recommender systems designed and implemented, we deploy these systems to real users of social network sites, and evaluate a variety of recommender algorithms online through user studies. Through these user studies we not only provide qualitative and quantitative comparisons of recommender algorithms, but also provide broader insights for designing recommender systems on social network sites.