Research On Personalized Hybrid Recommendation Algorithm Based On Improved Method

More information has appeared around people's life with the rapid development of Internet.The increasing information provides great convenience for people's life,making people get rid of the lack of information to a certain extent.But in the face of such a large amount of information,it is a perplexing question how to get exactly what people want.The existing search tool also appears powerless in front of such vast information,but the emergence of the recommendation system brought good news for people's individual needs,and makes information more efficient and convenient.This paper analyzes the commonly recommendation algorithms such as collaborative filtering,content-based recommendation,graph-based recommendation and association mining.Although these algorithms are applied in different fields,they still have problems such as inadequate adaptive ability and lack of personalized ability.Based on this,this paper attempts to improve the existing algorithm,and makes the improved algorithm mix according to certain rules.After analyzing the basic features of the MovieLens dataset,this paper uses the improved personalized recommendation algorithm to verify with this dataset.The validation results show that the recommendation's accuracy and coverage are superior to the traditional algorithm.In terms of algorithm improvement,this paper combines the project recommendation,association mining and the recommendation based on weighted two graphs to enhance the recommendation's effect.Tradition based on the recommendations from the project only used the rating data,did not consider the item's attributes.Such as for movie,it's properties can be the actor,director,film type,heat,and the quality of the film.Therefore,to improve the effect furthermore,the article mixes the content-based recommendation and the attributes of the content-based recommendation are added differents weights.Then mixes the similarity based on content and the data of score respectively.In order to solve the user's cold startup problem,this paper incorporates the recommendation algorithm of association mining.Because this recommendation based on association mining is based on finding frequent item-sets and setting the confidence.So it usually recommends the products that they are popular and we can solve the problem of the user's cold start better.The recommendation based on graph algorithm was proposed as a new recommendation system of branch of research in recent years,the user and the project are designed for the network nodes.In the traditional diagram,the lines connecting the user and goods don't have weight assignment,so it cannot reflect the user's subtle mind response to the project.In this paper,we give the weight on the rates that users given and combines with the above two algorithms to improve the accuracy and diversity of recommendation system.At the same time,this paper improves the similarity of the traditional collaborative filtering calculation,making the algorithm more accurate after it is improved.In the end,the paper also considers the problem of the user's interest will decay as time goes on.So we add the model of time index correction,which improves the accuracy furthermore.