Research On News Recommendation Based On Fusion Of Content-Based Recommendation And Collaborative Filtering

In the news recommendation system,the recommender results of content-based recommendation have the advantage of personality.But this method exist the lack of diversity that recommended results will only contain the news similar with user's former read.Collaborative filtering doesn't contain the problem of diversity,but needing several hours to accumulate user's clicks,hence causing cold-start problem.In addition,collaborative filtering can't satisfy user's individual necessary,which has been called as lack of personality.The hybrid method of collaborative filtering and content-based can accomplish both the need of diversity and personality,but still faces the problem of cold-start.To solve these problems,a new method of news recommendation based on fusion of content-based recommendation and collaborative filtering was proposed.This method could accomplish both the need of diversity and personality,and avoid cold-start problem efficiently.Firstly,this paper introduced background,current condition and general methods of news recommendation,and analyzed the lack of general methods.Secondly,the content-based method was used to find the user's existed interest,and the collaborative filtering was used to find the user's potential interest.Then,fusion user interest model which contain both the property of diversity and personality was constructed.Lastly,recommender results were generated by calculating the similarity between user interest model and candidate news feature model.When finding user's existed interest,this paper proposed an approach for the construction of existed user interest model with time weight,considering the relevance between user's interest and the time.When finding similar user set,this paper proposed a method to hybrid behavior similarity and content similarity to find the similar user set,considering the lack of accuracy of similar user set.This paper uses the data set provided by DataCastle,chooses F-measure and Diversity as assessing index,and uses content-based method,collaborative method and hybrid method as the baseline to make experiment.The F-measure and Diversity increased 34.4%and 38.2% respectively when fused method compared with content-based method.The F-measure increased 8.6% when fused method compared with collaborative filtering.And the F-measure and Diversity kept the same when fused method compared with hybrid method,but fused method shouldn't waited enough clicks so solved the problem ofcold-start.These results implied that the method proposed by this paper could overcome the lack of diversity and personality,and could avoided cold-start problem efficiently.