Link Recommendation in Twitter Networks

Twitter is a popular on-line platform for social networking and information sharing. Twitter users can follow other users to receive messages, or tweets, from them. However, given the limited attention and time, most users would like to follow only the most relevant users. Due to the huge number of users in Twitter, recommendation algorithms are needed to help users automatically discover new interesting users to follow.

Many existing link recommendation algorithms for social networks are developed with focus on the link structure. The simplest example is to recommend the most popular users with the largest number of connections. Other common algorithms first weigh each link by some importance score, and rank the nodes according to the sum of importance scores of their links, e.g., the PageRank algorithm. We generally refer to those algorithms as "popularity" or "weighted popularity" based algorithms.

Yet, unlike other social networks such as Facebook, besides to establish social connections, many Twitter users follow other users to receive information interesting to them. Hence, it seems attractive to exploit the similarity between Twitter users for recommendation, i.e., to recommend other users similar to the followees already followed by the follower, or to recommend other users who are similar to the follower. There are two general classes of algorithms for this purpose: the content-based algorithm and the collaborative filtering algorithm. The content-based algorithms match user interests by directly analyzing the texts of tweets, whereas the collaborative filtering algorithms such as matrix factorization technique learn latent user interests from the user feedbacks, e.g., to follow a user or not.

We compare various popularity-based algorithms and similarity-based algorithms for Twitter user recommendation, and propose hybrid recommendation algorithms to exploit both popularity and similarity. Indeed, a recent study has shown that popularity and similarity are the two important factors that drive the growth of a variety of networks including the Internet and social networks