Exploring Social Network Information for Solving Cold Start in Product Recommendation

摘要

Cold start problem is a key challenge in recommendation system as new users are always present. Most of existing approaches address this problem by leveraging meta data to estimate the tastes of new user. Recently, social network has been becoming an integral part of daily life. Usually, social network information reflect users preferences to some extent, combining this kind of data would contribute to address the cold start problem. Existing approaches of this kind are either leverage relationships between users or utilize meta data such as demographic information. The huge textual information in social network has been neglected. In this paper, we propose a novel recommendation framework, in which the textual data in social network are used to improve the recommendation accuracy for new users. In particularly, both of new user's interests and items are modeled by mining the textual data in social network. Experimental results demonstrate that our approach is superior to other baseline methods in both precision and diversity.