Improvement of recommender systems using confidence-aware trust

Collaborative Filtering (CF) is one of the most successful recommendation techniques. Regardless of its success, it still suffers from some weaknesses such as data sparsity and user cold-start problems, resulting in poor recommendation accuracy and reduced coverage. Trust-based recommendation methods incorporate the additional information from the user´s social trust network into collaborative filtering and can better solve such problems. However in these methods the level of confidence in direct and indirect trust estimations is under question. In this paper, an innovative Confidence-Aware Trust (CAT)-based recommendation approach is proposed within the CF framework An evaluation is performed on the Movie Lens dataset. Experimental results indicate that the CAT approach outperforms existing recommendation algorithms in terms of recommendation accuracy and coverage.