Research And Application Of Video Recommendation Algorithm Based On User Social Network

With the rapid development of artificial intelligence technology in recent years,the recommendation system is one of the applications in the field of artificial intelligence and begins to integrate into people's daily life.It intelligently filters the content of the user from the user's interest and the mass user behavior data,and helps people to make better reference and decision.In the early days,the search engines represented by Google and Baidu were displayed by user search content,but their recommendation was limited to the user's thinking.Unlike search engines,the recommendation system pays more attention to providing high quality content for users and building a bridge between users and information.Today,the Internet Co will combine them closely to help users better experience information services and improve user interaction satisfaction.QQ,WeChat,Weibo and other social platforms can not meet the social needs of young people,and traditional video websites lack interaction among users.So the video social networking sites such as A station and B station were born.The B station for example,by the end of September 2017,through the data obtained by the crawler B station preliminary statistics,the number of registered users has exceeded 200 million,the video submission number has reached 14 million 900 thousand.In order to increase the user's stickiness and achieve high quality video recommendation effect,the common recommendation algorithm is discussed in this paper.The user's preference is excavated from a large number of user information,video information and grading information,and personalized recommendation is realized.To some extent,a large number of search time and decision time are saved for the users,which greatly improves the practicability of the recommendation.This paper recommends the user's explicit feedback data for the user.Compared with implicit data,explicit data has stronger emotional color,including users interested and not interested in content,that is,positive and negative samples exist,but no need to construct negative samples based on data.Moreover,it is the mostclassic and most commonly used recommendation.In order to implement a complete video recommendation system,this paper has done the following research work:1.Data source acquisition: crawling "BiliBili" APP through a web crawler in a multithreaded way.The data source includes user information,video information,and scoring information.Among them,the user information contains the list of users' attention,which provides data support for the introduction of social recommendation algorithm.2.We discuss recommendation algorithm: memory based collaborative filtering algorithm,model-based recommendation algorithm,feature based content filtering method,association rule method and logarithmic likelihood ratio method.On this basis,the social network element is added,and an improved socialized recommendation algorithm is proposed.3.Experimental comparison of the algorithm: combined with the proposed and proposed algorithm,it is measured by the test index under the line.There are two kinds of commonly used offline inspection indicators: the root mean square error and the classification effect measurement method.They have realistic guiding significance from the two angles of prediction score and prediction correlation.4.Set up a video recommendation website: select the Python programming language,to ensure the consistency of the front and back frame language.The use of the latest Django2.0 Web framework has certain timeliness and research value.The use of flexible,stable,high quality Bootstrap to render the page,increasing the appreciability of the interface and the experience of the user.