Research And Application Of Recommendation Algorithm Based On Collaborative Filtering

Information is an indispensable part of the Internet resources.With the explosive growth of the Internet information,it becomes increasingly difficult for users to choose the information which is valueable to themselves.In order to promote the users' satisfaction,the economic profit,and the customer retention rate in the website,the personalized recommender system was born as a tool and a technical method to solve information overload problem.Different from other classic tools and technologies(such as search engines),the proactive recommender system automatically pushes recommendations to the user,without explicit request from him.Nowdays recommender systems are widely used in news,videos,E-Commerce and so on,but there are still issues such as the data sparsity problem and the cold start problem to be solved.Based on existing researches,the traditional recommendation algorithm only uses user-item rating data and ignores the potential impact through the items preference information,thus limits the recommendation quality.The thesis proposes a novel algorithm uses implicit similarity in preference relationships to address the above issue.First of all,it computes user interests distance according to the user attribution feature,and divides the users into different user groups based on the k-medoids algorithm.Secondly,it compares the item preference in different user groups,and gives the formal definition of the implicit similarity according to preference relationships.It introduces the items preference impact to the Probabilistic Matrix Factorization recommendation algorithm to improve the quality of the recommender system.The experiment result shows that the proposed algorithm improve the precision effectively.To solve the incorrect similarity computation problem caused by the data sparsity problem and the cold start problem in current collaborative filtering recommendation algorithm,it analyzes the availability of tag data to introduce the tag similarity,and incorporates the similarity calculation method based on the rating data.The collaborative filtering recommendation algorithm incorporates the tag information is proposed and a new similarity calculation method is given to search the nearest neighbor set of the target item.It predicts the target user preference about the item and generates more accurate recommendations according to the history behavior record.The experimental result shows that the proposed algorithm utilizes the additional information of the item efficiently and enriches the similarity computation data source and improves the precision of the similarity computation and meets the personalized recommendation demands.According to the above research,it develops the theme recommender system TSRS based on the mobile environment in engineering,the system overall design is introduced,and development environment and the final result is given.