Users' behavior session: Integrate passive feedback into microblog recommendation

Abstract

As one of the most popular social networks, microblog has been an important way for people to obtain information, meanwhile, the information overload problem is getting worse, which makes microblog recommendation become very important. Moreover, traditional recommendation methods cannot offer a good solution to this problem for the timeliness of microblog. Many researchers have make contributes to this problem, based on kinds of information, including users' interest, the history of users' behavior, and social relationships. However, most of these methods only use the positive feedbacks in users' behavior and treat these feedbacks independently. We consider that users' behavior should be persistent in a particular time range, which we called the activity session, then we define the passive feedbacks in sessions, and propose various contextual features for the activity session, and integrate these features to microblog recommendation by using both positive and passive feedbacks. Experimental results based on the real data of Sina Weibo show that, compared with the current popular recommendation methods, our method can achieve better performance. © 2014 Springer International Publishing Switzerland.