Preference Ontologies based on Social Media for compensating the Cold Start Problem

Abstract

Recommendation systems leverage future internet services to predict personalized recommendations for products, services, media entities or other offerings. Based on the research and development of the FIcontent 2 initiative, we introduce an approach to compensate Cold Start and Sparsity Problems by analyzing semantics of external textual data, in terms of comments from social networks as well as item reviews from product and rating services. Thereby sentiment analysis and semantic keyword extraction approaches are explained and evaluated by using preliminary implementations. The mined data is transferred into, so called, preference ontologies describing the users interest in automatic analyzed topics and subsequently mapped to the properties of items in order to calculate the associated recommendation value.