Research On Personalized Recommendation Method Based On Improved Collaborative Filtering

The development and popularization of Internet technology has greatly enriched the types and channels of people's access to information,and has also brought great convenience to people's work,learning and life.However,these vast amounts of data have already exceeded the limits that individuals can accept and produce the famous problem of "information overload",which seriously affects the efficiency of users to obtain effective information.Therefore,personalized recommendation system emerges as the times require.Since personalized recommendation system can help users extract effective information from a large number of redundant information through data mining technology,and then provide customized decision support and information services to users,so,in video,ecommerce,News and other areas of the system has been widely used.However,the recommendation accuracy and real-time performance caused by the sparse data problems are not good enough.Especially as the size of the data continues to expand,these problems become more prominent.Therefore,in this paper,a user-project hybrid collaborative filtering method based on radial basis function neural network(RBF neural network)is studied.The specific research work is as follows:(1)A user social network model and a method for constructing a user trust set are proposed.For the traditional collaborative filtering algorithm,the user similarity is calculated globally,which causes a lot of computing resources to be wasted,affecting the real-time problem of the system.The social network model is constructed through the user social information,the trust set is built by the model,and the similarity is built in the trust set to construct the neighbor.The way of collection changes the global calculation to local calculation,which improves the real-time response of the system.(2)A scoring prediction model based on radial basis neural network is proposed.For the traditional collaborative filtering algorithm,when calculating the prediction score,it only depends on the score of the neighboring users,ignoring the problem of the user's own scoring characteristics.The radial basis neural network is used to construct the scoring prediction model,so that the prediction score is closer to the user's real score and the data is alleviated.The error caused by sparseness improves the accuracy of personalized recommendation.(3)A user-project hybrid collaborative filtering algorithm based on radial basis neural network is proposed.Adjusting the scale factor combines two collaborative filtering algorithms to obtain the final prediction score,and improves the adjustment ability and recommendation accuracy of the personalized recommendation system in the case of data sparseness.(4)Based on the above research,the film personalized recommendation system based on Hadoop distributed computing platform web client is designed and implemented.Specifically,it provides service functions such as personalized customized movie recommendation list,added friends between users,movie category browsing,and movie rating evaluation for registered users in the system.