Study On Recommendation Algorithms Based On Collaborative Filtering Integrating Side-Information

With the popularity of computers and the rapid development of Internet,more and more information services bring convenience to users' daily life.Especially in recent years,the popularity of smart phones enables users to share or get information at anytime and anywhere.The information service makes a trend of exponential growth for the information resources on the Internet.The network filled with more and more information,would not lead users to find the 1 useful information quickly because of the massive data,which is also called the problem of information overload.In order to enable users to efficiently obtain the information that they need,the recommendation system represented by personalized recommendation technology emerges as the time requires.As the most successful application of personalized recommendation technology,collaborative filtering recommendation can extract users' potential interest only by their historical score data.It also has a series of advantages such as simple application and high accuracy.However,the traditional collaborative filtering algorithm has to deal with more serious problems because of the increasing amount of data and more complex data types in the recommendation system.The most difficult one is the problem of data sparsity.However,for some new recommendations cases at present,in addition to the users' history score information,we can also get the rich side information of users and items,which can provide important data support for describing the preferences of users and the attribution of items thus the opportunity is brought to alleviate the data sparse problem.In this paper,the detailed analysis and discussion on how to fuse the edge information into the collaborative filtering algorithm are carried out and the specific contents are as follows:(1)According to the data sparseness problem in collaborative filtering,this paper proposes a recommendation algorithm Tri-CF Recommendation Algorithm which can improve the quality of the side information.The algorithm combines the user-based collaborative filtering recommendation and the item-based collaborative filtering recommendation with the model-based collaborative filtering recommendation.By adding the constraining smooth terms for users and items based on the latent factor model,so that similar users or similar items would show similar latent features,and the experimental results proved that the improved algorithm is of higher recommendation quality than the three separate algorithms.(2)The paper explains the Tri-CF recommendation algorithm with three recommendation scenes in reality.In the traditional movie recommendation scenario,because there is no effective side information,we directly used user similarity matrix to calculate the similarity between users and movies,and it is brought into the Tri-CF recommendation algorithm.In the scene of image recommended,the image visual information is an important side information.In the paper,we extract the visual features of the image through the VGG16 network structure and the similarity between images is calculated based on visual features,and the similarity can be verified when brought into the Tri-CF algorithm.When we add the social information into the scene of movies recommendation,the social information of users is an important side information.In the paper,we calculate the degree of similarity between the user nodes in social networks through the "Degree Node Favorable Index",and then the similarity can be verified when brought into the Tri-CF algorithm.In the paper,we prove that the Tri-CF algorithm can be used as a general recommendation algorithm to effectively enhance the quality of recommendation through the experimental verification on three different recommendation scenes.