Selection of Trust Mechanism in Recommender Systems

Abstract

Recommender Systems (RS) have emerged as an important response to the so-called information overload problem. They enable users to share their opinions and benefit from each other's. Recommender algorithms are best known for their use on e-commerce Web sites to help users find products they would appreciate from huge catalogues. The products may vary from books (e.g., Amazon.com), movies (e.g., Netflix), photographs (e.g. Flickr.com), or web sites (e.g., del.icio.us)... The traditional collaborative filtering techniques are able to provide high-quality recommendations by leveraging the preferences of similar users. However, recent researches have suggested that the traditional focus on user similarity may not be sufficient. Additional factors, especially trust may have an important role when it comes to making recommendations.

In this thesis, we study the different algorithms and the use of trust to improve the performance of collaborative filtering recommender systems. Our evaluation on MovieLens dataset shows that the dimensionality reduction method that uses LSI/SVD technique helps in providing better quality of recommendations. Trust also has positive impact on overall prediction error rates, however, giobal trust metrics may not he appropriate for trust-aware recommender systems due to their non-personalized nature.