The Research Of Collaborative Filtering Recommendation Algorithm Based On Local-Nearest-Neighbors Slope One And Dynamic Experts

The endless rise of E-commerce websites leads to the rapid growth in the number of users and goods,which results in the difficulties for the users to find valuable information in the complex commodity ones.Thus the personalized recommendation system was born.It can accurately provide recommendation service to the users on the basis of their personalization features.Collaborative filtering recommendation algorithm is the most classical algorithm of the field.However,it still suffers the problems of data sparsity,cold start,scalability and the relatively low accuracy,which needs to be further improved.In this thesis,an improved algorithm is proposed to improve the accuracy and scalability of collaborative filtering based on the relevant problems of collaborative filtering recommendation algorithm and the latest research results.The theoretical research and discussion in this thesis were carried out from the following two aspects:First,in view of the traditional Slope One algorithm to improve recommendation accuracy problem,a Slope One collaborative filtering recommendation algorithm based on local nearest neighbor was proposed.In the classic Slope One algorithm,the linear regression model is used to predict the target project.However,some noise data are generated during the construction of the project score deviation table,which affects the performance of the algorithm.Slope One algorithm based on local nearest neighbor calculates the nearest neighbor user set for different recommended products to achieve the dynamic change of the neighbor set according to the different objectives of the project.The average deviation between items is further optimized and recommendations are generated according to neighbor user data of different target items.The experimental results show that the improved algorithm can promote the prediction accuracy of recommendation.Second,the collaborative filtering recommendation algorithm based on dynamic experts was proposed in this thesis to solve the scalability problem in the collaborative filtering algorithm.A dynamic expert database that changes with the change of the target projects is established according to the target projects,making the expertise areas of the experts fit the background information of the target projects.Through the calculation of the similarity between the active user and the experts in the database,recommendation is generated using expert opinion.Comparing experiments results show that the proposed algorithm greatly reduces the computation time and space complexity,and solve the scalability problem in the collaborative filtering algorithm in the case of maintaining relatively high prediction accuracy and recommendation precision.