On Solving Cold Start Problem in Recommender Systems Using Web of Data

Abstract

Data on the web has grown insanely large at a point it becomes unmanageable and very difficult to deal with using traditional tools. Hence the need for adequate tools to filter such enormous size of information and extract only the useful part has risen. Recommender systems are one of the de facto tools for such purpose. Varying from collaborative filtering to content-based filtering; their primary goal is to suggest the suitable items for the suitable users. However, due to the lack of information about both entities, especially the new ones, these systems may suffer from what is known as the cold start problem that prevents delivering appropriate recommendations. In this work, we propose a solution to overcome the two issues related the cold start problem, namely user and item cold start. The main idea is to use the web of data, a publicly available set of interlinked data and documents, to extract supplementary and useful information about new users and items which allows feeding the recommender systems with more relevant data. The proposed solution can be used an extension i.e. plug-in to an existing recommender system offering additional features to that system. The results of the experiments performed on the Movielens dataset are very promising and show the effectiveness of our proposal.