Research On Collaborative Filtering Recommendation Algorithm Based On User Interest And Item Popularity

In recent years,the use of e-commerce and multimedia short video data in my country has continued to increase,and the types of websites and products have grown rapidly.In this complex and massive data,how to efficiently filter information has become the current key research in the Internet service industry.Traditional search engines focus on static network technology,while ignore the research on user dynamic information changes.In order to balance the huge amount of data and complex user needs,a personalized recommendation system that connects information and user carriers came into being to help users extract keywords from the database,effectively alleviate the problem of information overload,and analyze it intelligently user needs.Based on the above analysis,the article has done the following research on the algorithm in the personalized recommendation system.The main work is as follows:(1)For the problem of sparse user explicit ratings,a revised user-item rating matrix is proposed.The traditional collaborative filtering algorithms can no longer solve the problems of sparse user rating matrix and timeliness of scoring data.Therefore,this paper uses the gradual characteristic of forgetting function to establish time threshold to classifies and modify user's explicit score,solves the problem of user score changes with time,reduces score errors,and further improves the problem of sparse score matrix.(2)For the problem that users' hidden behaviors are easily overlooked,a scheme based on interest modeling under tags is proposed.The existing collaborative filtering algorithm pays more attention to the user's explicit rating data,but ignores the mining of the user's hidden behavior,which limits the effect of the recommendation algorithm.The paper combines the characteristics of the dynamic changes of user interests and proposes a way to mine user hidden information based on user interests and tags,use the time and frequency of user tagging to measure the size of user's interest value,and weigh the role of long-term and short-term interests in the recommendation algorithm,solve the characteristics of the mining of user's hidden information and the dynamic changes of user interests,so that the personalized recommendation system have wider applicability.(3)Popularity bias is an important factor that lead to the decline of the accuracy of traditional recommendation algorithms.This paper proposes a scheme based on item popularity.By setting a popularity penalty value,the effect of popular products on the recommendation algorithm is reduced,and the impact of popularity differences on similarity calculations is reduced,thereby improving the performance and accuracy the personalized recommendation system.