Research On Collaborative Filtering Recommendation Algorithms Based On Tag And Trust Neighborhood

In recent years, the collaborative filtering recommendation technology has been widely used in the field of advertising, film, music, etc. However, in different application background, there are still problems need to be further solve for the traditional collaborative filtering recommendation algorithm interest preference model, similarity measure, neighbor selection rules, etc. Therefore, this paper faces advertising and movie recommendation and imports tag technology and the idea of trust neighbor, focusing on collaborative filtering recommendation based on user.The main research works of the thesis show as follows:1. We proposed a Advertising recommendation algorithm based on collaborative filtering with tag(ADR-CF_T). The algorithm introduces tag recommendation technology on the basis of the collaborative filtering advertising recommendation based on user. Through constructing search AD interest preference model called Q-K-A(Query-Keywords-AD) and using query page weighted comprehensive similarity, the search AD interest preference has a complete description and it ensures the accuracy of neighbor calculation. The experimental result of parameter adjustment, extensible authentication in KDD CUP 2012 track 2 dataset, and recommend quality comparison experiment shows that ADR-CF_T algorithm is not only effective and feasible, but also has been improved in the aspect of precision, recall, and F-measure.2. We proposed a Film Collaborative Filtering Recommendation Algorithm Based on Trust Neighbor(FCFRA-TN). The algorithm optimizes the traditional film collaborative filtering recommendation algorithm based on user using enhanced similarity, dynamic neighborhood selection method, and trust calculation model, which can not only correctly define the similarity between users, but also fully filter user neighbor set. The experimental result in Movie Lens dataset shows that what FCFRA-TN algorithm proposed in neighborhood selection strategy and trust calculation are valid. Comparing with other recommendation algorithms, it has a lower MAE and improves recommendation quality a lot.The research contributions of this paper include improving the traditional collaborative filtering recommendation algorithm based on user by optimizing interest preference model, similarity measure method, similarity description method, neighborhood selection strategy, trust calculation, etc. In the meanwhile, we applied it to the scenario of advertising and film recommendation. The experimental results show that the optimization of the proposed algorithm is effective and feasible.