Research On Short Video Recommendation Optimization Method With User Based Collaborative Filtering

The amount of information people faced every day is increasing massively for the reason of quick development of Internet.The main tools to help people find information of interest from the sea of data are categorized directories,search engines and recommendation technologies.Among them,recommendation technology has realized the transformation from traditional "people find information" to "information find people",which is widely used in online shopping,online audio and video,social networking sites and other fields.In recent years,with the support of recommendation technology,short video platforms such as ‘Douyin’,‘Kuaishou’ and ‘We Chat channel’have become one of the mainstream products of Internet applications,and whether the recommendation results of each platform meet the personalized needs of users is an important factor affecting their success.With more and more short video users,the daily data volume is getting bigger and bigger,and the personalized demand of users is getting higher,which makes the platforms have higher requirements on recommendation technology in terms of execution efficiency and recommendation quality.Therefore,this paper combines the characteristics of short video applications and conducts optimization research on the commonly used recommendation algorithm based on collaborative user filtering to achieve the better efficiency and precision of the recommendation algorithm.The main research contents of this paper include the following aspects.First,a user collaborative filtering recommendation algorithm process that adapts to incremental changes is proposed.After analyzing the impact of the data set size on the accuracy and time consumption of the algorithm,the short-term stability and computation frequency of similar users are analyzed,and an algorithm flow that adapts to incremental changes is given to reduce the impact of large data volume on the efficiency of the algorithm,while taking into account the requirements for the timeliness of the recommended content.Second,a user similarity algorithm considering user behavior time difference and video popularity is proposed.The user similarity is calculated by introducing the user behavior time factor and the video popularity factor.The experimental data show that the improved algorithm(TP＿User＿CF)has significantly improved the recommendation accuracy and MAP(Mean Average Precision)index.Again,an explicit and implicit feedback scoring method for different user behavior patterns is proposed.We analyze the characteristics of users’ explicit and implicit feedback behaviors,divide users into four categories according to their behavioral patterns,and propose an explicit and implicit feedback scoring method that can be applied to different categories,which solves the problems of difficulty in obtaining user ratings and large variability of user behaviors,and improves the sparsity of the user-project scoring matrix.Finally,a collaborative user filtering recommendation algorithm based on explicitimplicit feedback scoring(TP＿EI＿User＿CF)is constructed,which integrates the explicitimplicit feedback scoring method for different user behavior patterns and the user similarity algorithm considering behavior time difference and video popularity into collaborative filtering,and conducts comparison experiments with the traditional User＿CF algorithm and TP＿User＿CF algorithm.,and the experimental results show that the algorithm performs better in accuracy and MAP indexes and can effectively improve the recommendation quality.