Facet-based trend modeling for cold start of recommendation in social media

Abstract

Recommendation systems have been widely used in social media. One of recommendation tasks in social media is to provide relevant messages for users. Although many models have been proposed, how to make accurate recommendation for new users with little historical information still remains a big challenge, which is called the cold start problem. In order to address this problem, many models have been proposed, which use information of social media to improve the recommendation performance. However, lack of such versatility limits the successful application of these models. In this study, we propose an effective facet-based trend model to describe the trend interests of the entire user community in social media. Trend facet is the probability distribution of all users' preference to an attribute. In contrast to the general feature, the facet stems from the users' history and captures the interests to the attribute in social media. We evaluate our models in the context of personalized ranking of microblogs. Experiments on real-world data show that trend modeling can alleviate the cold start problem significantly. In addition, we compare the performance of user modeling and trend modeling, and find that user modeling outperforms trending model and the impact is slightly negative when combining trend modeling with user modeling.