Research On Accurate Recommendation Strategy Of Personalized Recommendation System

The rapid development of the Internet has provided users with a wealth of information resources and satisfied their needs for information in the era of big data.However,users' ability to collect,filter,sort and summarize information is limited,especially when dealing with a large number of data;it becomes extremely difficult for users to obtain available information from it.The emergence of personalized recommendation system can not only solve this problem,but also improve the coverage of user retrieval information and formulate personalized recommendation content.Among them,the application of the recommendation algorithm is an important prerequisite for the realization of the above functions.Collaborative filtering recommendation algorithm is widely used in various fields because of its simple implementation method,strong model scalability and good recommendation effect.However,such algorithms have problems such as cold start,unreliable user rating data,and single application scenarios of recommendation algorithms.In order to solve these problems,most of the current studies focus on the improvement of algorithms.This thesis attempts to improve the recommendation accuracy by comprehensively using the classic algorithms without changing the classic algorithms.An accurate recommendation strategy based on collaborative filtering is proposed which is named CFAR.In order to reduce the negative impact of untrustworthy scores on the recommendation results,this strategy designs a credibility processing method for user rating data;it cleans up the rating matrix according to the credibility of user rating data to remove untrustworthy ratings.Aiming at the user cold start problem,the strategy uses the K-means clustering algorithm to cluster the characteristics of all users,and recommends Top N favorite items of other users in the cluster the new user in to the new user.In view of the shortcoming of single collaborative filtering recommendation algorithm in practical application,this strategy chooses appropriate recommendation algorithm results for the existing user according to the size relationship between the number of items and the number of users,and the size relationship between the F1 values obtained by different recommendation algorithms.The experimental results on the Movie Lens dataset and the Film Trust dataset show that the CFAR strategy has improved recommendation accuracy compared with only using a single collaborative filtering algorithm.At the same time,it can eliminate the influence of untrustworthy scores.In order to better test the practicability of CFAR strategy,this thesis develops a simple food recommendation system.The system recommends foods that users may like by analyzing the user's characteristic attribute information and the user's historical rating data.The application results show that CFAR strategy can be effectively applied to the food recommendation system and can accurately recommend food.