Research On Hybrid Recommendation Algorithm Of Combining Collaborative Filtering And Content-based Recommendation

The rapid development of the network technology influences and changes human life.Advanced network technology is no longer a network to allow people to do something,but reflected in how people enjoying a better experience life on the network.This is the technology from simple to complicated,from coarse into fine change.The information overload problem seems to have affected the comfort of the people living in the network,however,the recommendation system bring the gospel to people who needs the personalized needs.Collaborative filtering and content-based is two main recommendation algorithm.Although the two recommendation algorithm is already being used in different areas,there are still some problems,such as lack of adaptive ability and personalized recommendations.In addition,advantages and disadvantages of two algorithms are complementary,but due to the constraints by non-text items with the feature extraction difficulty,content based recommendation algorithm is generally used only for the text of recommendation system.Therefore,Hybrid recommendation algorithm present by the mixing of the two kinds of'algorithms dosen't suit for every field.In order to improve the quality of recommendation algorithm and to play the advantages of the two kinds of collaborative filtering and content-based recommendation,the paper puts forward a hybrid recommendation algorithm.In the hybrid algorithm,collaborative filtering is the main algorithm and the content-based recormmendation is subsidiary algorithm which is used in the process of search the trusted neighbors.The final result is provided by these trusted neighbors.The innovation of the hybrid algorithm mainly includes:First,Introducing project heat to optimizing the pearson correlation coefficient.Second,putting forward the method of using labels as attributes of non-text items,building two dimensional interest model for users,and measuring the similarity of interest model.Third,according to the characteristics of interest model similarity formula,the paper proposed the method to solve the similarity weight coefficient by using the variance.The experiment shows that hybrid recommendation method can improve the recommendation quality.It is an effective method and it has the recommended advantages compared with two existing mixed strategy.The effective of every step of the optimization and the calculation method of the interest model are verified by corresponding sub experiment.The feature of items is extracted by their tags,so,the hybrid recommendation method has certain universality for non-text items recommendation.