Optimization of Collaborative Filtering Systems

A collaborative filtering system (CFS) makes recommendations to users via the similarity and proximity between the profiles and taking into account their historical valuations. In contrast to most of the CFS, which are based on the approach-based users, we adopt the approach based items to improve the quality of recommendation, this process seems flexible and allowed us to integrate other sources of information while making the calculation mode off-line, and then to improve performance and reduce the inconvenience of the lack evaluation, we used the semantic layer objects such as metadata and semantic relationships between items, finally we explored the technique LSI (Latent Semantic Indexing) to reduce the complexity of algorithm and identify the items most corollas. A set of real MovieLens test has been used for experimental tests.