Research On Information Fusion Recommendation Method Based On User Interests And Characteristics

With the emergence of internet and the popularity of applications in various fields,people were provided with a large amount of information,which greatly satisfied their requirements for obtaining comprehensive and convenient information.However,with the rapid development of network applications,the content of network information is continuously enriched,and meanwhile the number of information has rising exponentially year by year.Accordingly,when facing massive information,it is difficult for users to obtain key and valuable information effectively.Thus,a lot of time and effort have been put on information filtering and value identification,which means the convenience of the past but the burden of today.The emergence of the recommendation system not only solves a series of problems caused by information overload for users,but also helps enterprises to maintain customer relationships and improve sales capabilities.Based on the above reasons,the research and application of the recommendation system has received great attention from academic circles and industrial circles both at home and abroad.However,the current research and application of recommendation systems are still not mature enough.Problems such as sparse data,poor real-time recommendation,and poor recommendation are still common.Focusing on the scientific research and practical application of recommendation system,the paper analyzes the research status of the recommendation system at home and abroad,and sorts out the research results of the existing mainstream recommendation methods.It is found that user interest modeling,one of the basic and core issues of the recommendation system,has problems such as insufficient segmentation of interest and single information structure used for user portraits,which results in inaccurate user interest description.In the collaborative filtering recommendation method,the user interest is described as meticulously as possible so that users can find a set of neighbors with a more similar interest and thereby make a more accurate recommendation for them.Therefore,how to describe user interest more accurately has become a key issue needed to be solved urgently.Further,the effective measure of similarity between users is another core issue in the research of recommendation systems.However,the similarity between users is often described only through user interest information by existing studies.In fact,the user interest will change with time and therefore has great uncertainty.Hence,how to effectively integrate new user information and fully describe the similarity between users have become new urgent issues in the collaborative filtering recommendation research.This paper proposes a recommendation method based on user interest and feature information fusion on the basis of previous research,which is mainly around the above two core issues and the data sparseness existing in the recommendation system.Based on the user-based collaborative filtering recommendation method,the traditional user interest model is improved by introducing the user's long/short term interest and project rating,and the user's feature model is integrated to build a user's comprehensive similarity model.Based on this model,we recommend neighboring users as the target user.The main research content and innovation of this article are summarized as follows.(1)For the data sparse problem in collaborative filtering algorithm,this paper changes the User-Project score matrix into User-Attribute preference matrix through item attribute extraction,which reduces the matrix sparseness to some extent.(2)For the issue of describing user interest more accurately,on the basis of subdividing the project attributes,this paper analyzes the user's historical behavior,comprehensively considers the stability of long-term interest and the real-time nature of short-term interest and provides users the distinction with a long-term and short-term interest in project attributes.On this basis,the differentiated user interest attributes are merged with the project scores to build a user interest model which introduces the user's long-term interest and project ratings.(3)For the problem of how to accurately measure the similarity between users,this paper fuses the user feature similarity information based on the user's similarity study and then proposes a new user's comprehensive similarity model.Scoring prediction and Top-K recommendation are given based on this model for the product that the target user does not pay attention to.In this paper,a recommendation method based on user interest and feature information fusion is implemented through MATLAB tool.We validated the proposed recommendation method with 100,000 rating data of 1682 movies from 943 users in the ml-100 k public data set.It is found that this method shows a higher recommendation accuracy than the traditional method after comparing the average absolute error MAE and can solve the data sparse and the cold start problem to a certain extent.