Research And Implementation Of Accurate Friends Recommendation System Based On Social Pictures

Human society is composed of complex social networks,and people communicate and realize their own values through social communication.The rapid development of communication network and the popularity of mobile devices make social in network become an important part of people's lives.Making friends on the Internet can enjoy more mystery and is no longer limited by physical distance.It not only solves the problem of communication between acquaintances cannot communicate frequently because of the distance,but also enables people to meet more strangers to expand their social scope.Therefore,a large number of social applications have emerged,how to more accurately recommend friends to different users between different applications is an effective way to improve user stickiness and improve user experience.The friend recommendation algorithm can solve the pain points of various social applications and become a popular research direction.Generally,social application recommend friends is based on the number of mutual friends,the identity information of the user's occupation,geographic location,interest tag,and other factors.However,a friend recommendation method,which only considers a single factor,cannot meet user's needs,and it is difficult to comprehensively obtain the user's dynamic friend preference only by analyzing the user's social network and fixed tags.Therefore,this paper proposes a friend recommendation algorithm based on dynamic information to solve the above problems,which builds a demographic-based friend recommendation module based on the user's gender,occupation and other identity information,and the user's geographic location information,then calculates the similarity based on the Identity information and geographic location information.Two factors considered in the dynamic information-based recommendation module are the user's most recent interest in geographic location information and the user's social status.Pictures in social status can express more information and content,therefore,the recommendation module based on the dynamic information selects the interest location information in the user check-in and the social picture information in the social dynamics to constitute the user's interest vector.Use this vector to build a user-interest matrix and scored,then calculate the similarity between users based on the collaborative filtering idea to obtain the nearest neighbor set.In the recognition of user social pictures,adopted SSD target detectionalgorithm with high recognition accuracy and high speed to obtain the information contained in the social picture and generate the user's interest tag.Finally,a mixed recommendation method is used to linearly fuse the recommendation module based on demographic and the recommendation module based on dynamic information,and calculate the similarity for the friend recommendation.In this paper proposes a friend recommendation algorithm based on dynamic information and designs and implements a friend recommendation system based on this algorithm.using the data set of Sina Weibo users to conducte verification test,through comparing the accuracy and recall rate and F values,proves that the friend recommendation algorithm based on dynamic information compared with single factor recommendation algorithm has better recommendation result,verify the feasibility and advantages of the proposed algorithm.