Research On Personalized Video Recommendation Method In Social Network

With the increasing maturity of social networks and the rise of short videos,social networks are flooded with a large number of short videos.Video viewing has also become an important service in social networking sites.As a result,new forms of social media such as video and social media are rapidly emerging.Social networks such as social networks have emerged.In social networks,users mainly socialize with videos and text descriptions.Different from the traditional video media,the user's feedback operation in the social network not only can be rated,but also can be praised,forward and comment.Based on this,how to find videos that users are interested in in massive data and increase user satisfaction are currently hot research topics.Video recommendations in social networks are different from traditional recommendations.The main ones are:(1)there are more hidden information in social networks that can be used for recommendations.The user data in the social network includes not only the user's comments,ratings,and other information on the video,but also the social relationships of the user,such as the person user is concerned and his fans.The user may be more interested in the video posted by his friend.The video data contains more information about the content,the number of forwards,and the number of plays.so how to use this information is crucial for the recommendation system.(2)The large amount of data in the social network and the high degree of sparsity make the video recommendation unable to directly adopt the traditional recommendation method.therefore,this article is devoted to the research of personalized video recommendation based on the social network.Although the research of video recommendation methods based on social networks is very hot,there are still the following disadvantages:(1)The information of the video and user are not fully used,there is no effective solution to the problem.For example,the data transmission and other information of the video have not been taken into account and the sparseness of the data and the large amount of data have not been given a good solution.(2)The recommendation accuracy for the user in the social network is low and an effective method have not been given for this problem.therefore,we are aiming at the above characteristics of social network recommendation and the deficiencies in the current research stage based on the existing recommendation theory and technology,The main work of this article includes:(1)Built a video recommendation model based on social networks.The model includes a data acquisition module,a feature extraction module,a recommendation module,Gradient optimization recommendation module and test module.The data acquisition module mainly solves the problem of the data acquisition and divides the data into training data and test data.The feature extraction module analyzes user and video data information and extracts three characteristics of user characteristics,social network and video information features.The recommendation module refers to content recommendation based on user preference,recommendation based on user collaborative filtering and recommendation based on social network service,otherwise it calculates corresponding preference degree.The gradient optimization recommendation module refers to measuring three recommendation results and establishing a learning ranking model and using the gradient descent method to obtain the optimal weight.Finally,the comprehensive high score of the video should be recommended to users according to the optimal weight.The test module is to test the effectiveness of the algorithm based on the trained model and evaluate the recommended results.(2)In this paper,we analyze three characteristics such as user characteristics,social networks,video features and describe the calculation methods for each feature.user characteristics mainly analyze user preference features adopt traditional content-based recommendation algorithm according to history browsing history in user social network.we extracts user's preference keywords calculating keyword weights and establish user's preference.user can be sparse subspace clustering to solve the problem of large data volume and data sparseness by using feature vectors.The social network feature analyzes the characteristics of the user's social friends and adopts a collaborative filtering method to measure the similarity between users by integrating the trust and similarity between users and predicts the user's rating for the recommended video based on the user's close friends and potential similar friends.Video features is refering to features that contain video content information and video influence.For video information,we extract video titles and keywords in the content such as actors and directors to create video vectors and compare them with the user's preference documents to obtain the user's preference for the recommended videos.video influence is referring to the popularity of the video.It measures a characteristic of video influence according to data such as video likes and at the same time it solves the influence of the cold start problem on the recommendation effect.(3)The video information features,user preference features and social network service features are analyzed and extracted.The video information feature refers to a feature that describes a video information keyword.for the characteristics of the user,it means the user preference feature is mainly analyzed and the user's preference keyword is extracted according to the historical browsing record in the user's social network.The social network service feature refers to the similarity,interaction,trust characteristics,and popularity characteristics of the video among social users.The similarity between users refers to the similarity of interests between users.For the degree of interaction,it refers to the characteristics between users who have had interactive behavior.The degree of trust refers to the social relationship with the user such as paying attention to users,fans.The video popularity feature analyzes the popularity characteristics of video in social networks.Video popularity refers to a feature that measures the influence of video according to the number of video viewers in terms of the number of times the video is played,and at the same time it can be used to solve the user's cold start problem.(4)A personalized video recommendation algorithm based on social network is presented.Firstly,according to the established user's preference feature vector,the similarity is calculated with the video feature vector to be recommended and then the video and similarity value with high similarity are selected.According to the idea of collaborative filtering,the top N users with high similarity values of the target users are searched and the historical browsing records are analyzed,the video sets to be recommended are obtained,and the scores are predicted,and the to-be-recommended video sets are obtained according to the ranking.According to the characteristics of the social network service,the user's interactive user and the trusted user are respectively used to calculate the score value of the user's recommended video,and then the random forest algorithm is used to calculate the user's preference value for the recommended video through the video popularity and the user's age and gender characteristics.The score value is finally recommended to the user by using a linear weighting and SVM idea based on three different social network service feature values.In order to comprehensively measure the three recommended results,this paper establishes a learning ranking model and adopts the method of gradient descent to measure the hinge loss function,so the optimaled weight is to be used to calculates the user's comprehensive score for the candidate video and The video with high score is recommended to the user.To verify the validity of the recommended method given in this paper,the experiment selects the accuracy and recall rate as the evaluation index of the recommendation result.Using the real data crawled from the Tencent video data,compare it with the existing methods of recommendation and the different features considered.The experimental results show that compared with the existing recommendation method and the single feature recommendation method,the recommendation method based on the social network given in this paper is better and it can be more effective for the user to make video recommendations,improve the user experience and meet the user's needs.