Using multi-features to partition users for friends recommendation in location based social network

Friend recommendation is an important feature of social network applications to help people make new friends and expand their social circles. However, the user-location and user-user information in location based social network are both too sparse which contributes to a big challenge for recommendation. In this paper, a new multi-feature SVM based friend recommendation model (MF-SVM) is proposed which regarded as a binary classification problem to tackle this challenge. We extract three features of each user by new methods respectively. The kernel density estimation and information entropy are used to smooth the check-in data and highlight the activity level of users to extract spatial-temporal feature. Then the social feature is extracted by considering the diversity of common friends. After that a new topic model improved by LDA is proposed which both considers user reviews and corresponding service description to extract textual feature. Finally, these features are used to train the SVM and whether the users have a friend link can be predicted by our model. The experiments on real-world datasets demonstrate that the proposed method in this paper outperforms the state-of-art friend recommendation methods under different types of evaluation metrics.