Semi-supervised Tag Recommendation - Using Untagged Resources to Mitigate Cold-Start Problems

Christine Preisach, Leandro Balby Marinho, Lars Schmidt-Thieme

Abstract: Tag recommender systems are often used in social tagging systems, a popular family of Web 2.0 applications, to assist users in the tagging process. But in cold-start situations i.e., when new users or resources enter the system, state-of-the-art tag recommender systems perform poorly and are not always able to generate recommendations. Many user profiles contain untagged resources, which could provide valuable information especially for cold-start scenarios where tagged data is scarce. The existing methods do not explore this additional information source. In this paper we propose to use a purely graph-based semi-supervised relational approach that uses untagged posts for addressing the cold-start problem. We conduct experiments on two real-life datasets and show that our approach outperforms the state-of-the-art in many cases.