Evaluating user profiles in a micro-blog environment for the purpose of recommendation is a challenge because of the limited amount of information available in short text. Zheng et al. present the framework of the neighborhood user profile (NUP) and experimental results, utilizing NUPs in addition to the normal user profile.
While the amount of text allowed in a micro-blog system is usually limited, a typical micro-blog system does have “following” or “being followed” relations among its users. NUP makes effective use of these two relations in assessing user interests. NUP uses two key constructs: resource perception relationship (RPR) and follow perception relationship (FPR). RPR between two users is defined as the similarity among the posts from the two users, while FPR between two users ui and uj is defined by the set of users following ui or uj and the set of users being followed by ui and uj. Extracting information from the neighbors using RPR and FPR allows NUP to compute profile similarity among neighboring users; thus, meaningful recommendations can be made.
Two real-life, 20-day datasets from Sina blogs are used to evaluate the performance of NUP. The first is from December 2011, which involves 85 users and 1,247 followees. The second is from April 2013, which involves 75 users and 4,013 followees. The information retrieval measures of recall and precision are used to assess the performance. In all measurements, NUP shows superior performance compared to other contemporary measures.
The paper is well written and self-contained. It should be of interest to anyone who works in the areas of user interest discovery, user profile digests, and recommendation in micro-blog systems.