Personalized Recommendations Research Based On User Behavior Analysis And Hybried Strategy

Information resource has become an important part of society. With the rapidincrease of information resources and the efficient access of information, excavatingmeaningful information from these information also becomes an emphasis. In order toimprove users’ experience in access the website, reduce users’ information browsingtime and improve customer loyalty, personalized recommendation system and itsrelated research generate. At present, personalized recommendation system is widelyused in the field such as video watching、music playback、news browsing, andelectronic commerce. This article mainly discusses the processing of structured datain the recommendation system, user model building, the recommendation algorithmand its mixed strategy. Hybrid recommendation method is used in solving the problemsuch as new item recommendation, sparse data user recommendation, and finallyimplements the video recommendation system.In the personalized recommendation system, the text description data of video isgot in the way of web crawler, and then the video text data is cleaned and structured,and structured description data of the video will be generated ultimately,which playsan important role in the recommendation system. In the analysis of user behavior,users’ preference value of the project is generated by analyzing user behavior data.According to users’ preference value of the project and structured data of the project,the content-based preference features is generated. At the same time, this paper usestriples data model to describe the user’s project data, and uses vector space model toconstruct the users’ content-based preference model.This paper adopted three recommendation algorithms and two results mixmethods to generate user’s recommendation results when constructing arecommendation system. The recommendation algorithm include three algorithms:content based recommendation algorithm, user based collaborative filteringrecommendation algorithm and the hybrid recommendation algorithm based onassociated projects and similar content. In the design of collaborative filteringrecommendation algorithm based on users, different similarity measure methods arecompared, and finally used the extended cosine similarity to calculate the similaritybetween users, which increased the recommend result. In the design of content basedrecommendation algorithm, used weighted cosine similarity to calculate the similaritybetween users and video data, through the study of the optimization of weights andimproved the effect of the recommended. In the design of the hybrid recommendationalgorithm based on associated projects and similar content, both similarity andcorrelation between content of the project were considered comprehensively. Thesimilarity between projects was calculated by the similarity between project’s structured data, and the correlation between projects was mainly obtained by theanalysis of the data of users’ access to the project. Hybrid recommendation algorithmneed to mix the similarity and correlation between projects together to get the hybridsimilarity. In the recommended phase, the most similar projects with user’s historicaldata will be recommended to the user.In the mixed phase, user data will be divided into active user data and inactiveuser data according to the user activity. For active users, results of therecommendation algorithm based on contents and the collaborative filteringrecommendation algorithm based on user’s recommendation will be mixed, solve theproblem of new item recommendation under the premise of ensure accuracyrecommended results. For inactive users, hybrid recommendation algorithm based onassociated projects and similar content will be adopted. Finally the effectiveness ofthe hybrid strategy is verified by experiment.