Based On The External Index Of The Network Short Video Recommended Method Research

With the advent of the mobile Internet era, people contact the infinite expansion of information content, unlimited channels. Short video is presented in front of the user as a new media form and quickly has a huge user size. Short video has a higher level of information, including richer information and because of bandwidth costs, people's consumption habits and pace of life, short video is more easy to accept and spread. At present,the Internet has a large number of short video information, these short video to the user has brought serious information overload problem, to the short video operators, video editing has also brought great challenges.Therefore, short video recommendation technology has been paid more and more attention and become an important research topic.The traditional recommendation process is often recommended by collaborative filtering or based on the content to the user, but these methods have some limitations. Short video spread quickly, it has a strong timeliness and thermal search. Different recommended methods can not recommend high attention, high heat short video. In this paper, the external index feature is added on the basis of extracting short video characteristics, and a short video recommendation method based on exponential feature is proposed. This paper mainly carried out the following three aspects of research work:(1) The characteristics of short video are analyzed, and the dominant and recessive features of short video are extracted. This paper studies the feature construction method for short video recommendation, and proposes a short video subject feature construction method which integrates external exponential characteristics.(2) This paper studies the short video recommendation method based on two categories. In this paper, the short video recommendation is ed into a two-class problem,and three different classification algorithms are used to achieve short video recommendation: factorization,gradient iterative decision tree and logical regression. In this paper, the short video recommendation method of three different classification algorithms is described in detail, and then the performance of different classification algorithms in short video recommendation is verified by experiments.The experimental data of this paper comes from a well-known domestic video aggregation APP log, on the basis of this experiment and comparative analysis. Firstly, the recommended performance of the three classification algorithms used in this paper is verified. The experiment shows that the performance of the factorizer is the best. Secondly, the experimental results show that the dominant features, hidden features and exponential characteristics of short video are effective features. Moreover,the short-video theme feature that combines the exponential features has the greatest effect. Finally, this paper also gives the use of LDA to build the theme features, select the number of subjects in the experiment.