Research On The Method Of User Behavior In Social Networks

Micro-bloggings, like Twitter and Weibo, have been widely popular and influential platforms for information diffusion and social interactions. Selecting a suited person to mention on the Micro-blogging network, expressed as @username, is a new aspect of recommendation system which carries much importance to promote user experience and information propagation. People would like to mention their friends or celebrities to promote products, report new events, share opinions or bring up questions. We address this problem by proposing a novel modified collaborative filter algorithm with the voted weight. Specifically, we calculate the similarity between the target tweet and the historical tweets while every historical tweet would vote for every candidate with a different weight, and eventually the approach returns top k users with high scores to the target tweet as recommendation.Specifically we regard information propagation within this problem as the vitality, reach and effectiveness of tweet message. With learning-to-rank framework, we propose our model, named as Personalized Mention Ranking, to find out who has the maximum capability and possibility to help diffusion by ranking their scores. In our work, we extract a wide range of features, such as tag similarity, text similarity, social influence, interaction history and name entity.Experimental results show that our approach outperforms the state-of-art algorithm. Traditional recommendation model, such as content-based recommendation and Collaborative Filtering recommendation, doesn’t work as well as ranking model. Concretely, ranking model represents to be more suitable than regression model in this @ recommendation issue and results of listwise models are superior to results from pairwise and pointwise’s.