An Online Video Recommendation System Based On User's Play Sequence

With advent of the internet and mobile Web 2.0 era,the video resources and users have grown in size.but with the ensuing serious information overload problems.For the video service platform,the personalized recommendation system can effectively obtain the user's interest preferences so that the users can get targeted and precise delivery.For the users,the personalized recommendation helps them to solve the problem of information redundancy.Therefore,as a win-win strategy for users and platform,the personalized recommendation system not only enhances the user's dependence on the platform service but also greatly shortens the time spent by users in obtaining the video resources.Under this background,this thesis makes a deep research on the personalized recommendation system,and proposes a personalized recommendation algorithm Seq4 Rec based on the current natural speech processing technology.The Seq4 Rec algorithm classifies the user's history video playing sequence into word sequences in natural language,processes the video ID sequence by the natural language processing algorithm and further obtains the corresponding video vector,so as to map the video without the feature and label information into several dimensions Video vector.However,it is worth noting that a single video vector has no practical significance in each dimension,and only relative information between video vectors is of practical value.The personalized recommendation algorithm proposed in this paper generates thousands of personalized video recommendations for each user according to the user's video viewing history and the similarity between videos.In this article,we first model the video using the current state of the art natural language processing algorithms and then make recommendations to users based on this,in conjunction with traditional collaborative filtering algorithms.According to the relevant principle of Seq4 Rec algorithm,experiments were conducted on the de-sensitized user historical play behavior data provided by Tencent Video.In the experiment,we first clean the irregular data and construct the algorithm training set according to the user's playing time sequence of eachvideo,and then use the skip-gram model in NLP to model the data and then make the user's history play video Clustering and further calculating the user's interest distribution matrix,and finally combining the user interest and viewing history to generate personalized video recommendation,and the experimental results are respectively compared with popular recommendation,user-based collaborative filtering and article-based collaborative filtering recommendation algorithm The results of a comparative analysis.Experimental results show that the proposed Seq4 Rec algorithm results significantly better than the contrast algorithm.