A Fused Method for News Recommendation

In news recommendation system, the content-based method face the lack of diversity that recommended results will only contain the news similar with user's former read. The hybrid method face cold-start problem that the candidate news must wait enough click to recommend. We propose a fused method which is capable of handling the diversity problem of content-based and cold-start problem of hybrid approach. We use content-based method to extract user's existed interest firstly. Find out similar user set and predict target user's potential interest by collaborative filtering secondly. Fuse existed interest and potential interest to obtain the final user profile the third. Generate the recommended results after comparison the similarity of candidate news and final user profile lastly. This approach results in an offline news set and compared Diversity and F-score with other recommendation method. It decreases the lack of diversity compare with content-based approach and avoids the cold-start problem compare with hybrid method.