A novel relationship strength model for online social networks

Abstract

One of the key foundations of personalized recommendation in a social network is the relationship strength between social network users. The improvement for recommendation accuracy is mostly tied to the precise evaluation of the relationship strengths. With most of the selected factors affecting the relationship strength between users are too simple, the existed researches show low accuracy in calculating the strength, especially those factors related to topic and indirect links. We propose an online social networks users relationship strength estimation model which incorporates topic classification and indirect relationship. We adopt K-means clustering method using ABC algorithm to cluster all the interactive activity documents and calculate the correlation between clusters and activity topic name. After that, we compute the relationship strength between users which belong to the same topic on top of the user profile and interaction data. To accomplish this we employ a language model based on sentiment classification approach and take similarity, timeliness, and interactivity into account. We conduct experiments on two microblog datasets and the results show that the proposed model is promising and can be used to improve the performances of various applications.