Video Recommendation Based On User Preference

With the development of information technology and Internet applications, thehuman society is striding from the information lacking age to information overload one.Traditional search technologies, as in the Internet information platform which isrepresented by video related service, cannot fulfill the users’ content for personalizedservice. Active analysis of users’ behaviors, to locate users’ preferences and unearth theinformation resources which conform to users’ individual demands is needed by theresources platform in the personalized Internet age. The personalized recommendationtechnology rises no other than to this challenge.In video resource system, the popularity of the videos, determined by users’different watching behaviors, presents as fat-tailed distribution, i.e., most-viewed videosare only a few. Majority of the videos are less favored. Moreover, users’ preferences arenot always fixed. By making advantage of users’ behaviors information, further analysisthe features and rules of users’ preferences which change s over time can be carried out.Meanwhile, users’ accumulated rating behaviors could manifest some kind of ratingtrend, like some demanding users or poor video quality; they tend to a higher ratingvideo. Excluding the effects of above-mentioned objective facts to the recommendationresults, the traditional video recommendation technology has not do well in this respect.This article gave a detailed introduction of video recommendation system and itsessential technologies. Against this background, two core issues of recommendationtechnology, Top-N recommendation and score predict recommendation, are mentioned.The improvements of recommendation algorithm modeling aim at the traditionalrecommendation technology have also been stated.First, this article presented users’ preferences modeling strategies whichcomprehensively make uses of the users’ implicit and explicit feedback. Then itproposed a collaborative algorithm, relied on users and items’ collaborative filtering,which can be applied to Top-N recommendation. The popularity weight of video and thechange weight of users’ preferences are inserted into users’ collaborative filteringalgorithm. The performance of three algorithms, on the ground of user-video binary related matrix model, is compared in the experiments. The experiments showed that thestrategy to take the popularity of the video and users’ preferences into consideration willeffectively enhance the quality of recommendation.Second, briefed on the existing solutions to the score prediction ofrecommendation system that based on the sparseness of the users’ rating data, especiallydwells on the representative algorithm, collaborative algorithm which leans againstmatrix decomposed model, and proposed an improved model inserted with bias terms.Finally, by the experimental analysis, verified that the improved matrixdecomposed model can effectively improve the accuracy of the score prediction.