| In the study of social computing, user behavior analysis is mainly aimed at the history behavior of the observed object and predict his future behavior, this study play a key role in the social computing applications. Retweeting behavior of microblog is a kind of high frequency behavior of user behavior, and the most important method of information dissemination on microblogging platform. The research on users’ retweeting behavior prediction is meaningful to understand information dissemination, hot event detection and personalized information recommendation.In this paper, we fucus on users’ retweeting behavior prediction. We simulated users reading habits when retweeting on microblogging, analysis the factors which influence the retweeting behavier from tweet, user, and the relationship betweet them. We use these factors as features, and a set up a classification model to predict which tweet a user is going to retweet. We performed a featrure combination experiment to verify the effect of all kinds of influence factors, among them, User history behaviers show the interest of users, in this paper we put forward two mathods to use the relationship between user history and tweets: the cartesian product of words and the similarity calculation based on topic model. The experimental results show that these two kinds of methods both have effects on users retweeting behavier prediction. When using the method of cartesian product, we got the best prediction result, the F1 value is 71.33%. We tried to add sentiment feature into the mathod based on topic model, which can also take effect, the predicted results of F1 value is 69.97%.We also present a new problem of predicting users retweeting sentiment in this paper, and put forward a preliminary solution. We label the sentiment of retweets automatically, and use the character of stable emotion on retweet to predict the retweeting sentiment. The experimental results show that our mathod achieved good results, the accuracy is 64.11%. |
PDF Full Text Request