Research On Social Recommendation Algorithm Based On User Preference

With the rapid development of the Internet,a large amount of data is generated on the network all the time.In order to enable users to quickly obtain projects of interest,the recommendation system came into being and is widely used in various application platforms.With the close combination of people’s online and offline life,social recommendation has become a research hotspot in the recommendation system,which has the value of theoretical research and practical application.The main research of this thesis is to recommend users with the help of users’ social network information and social tag information of users and projects.At present,most recommendation algorithms based on social networks assume that users’ social networks are homogeneous,treat the preferences of users and friends equally,and ignore the possible inconsistency between users’ and friends’ preferences in different projects.In the recommendation algorithms based on social tags,most of them do not consider the user’s attention to the tag,but only take the tag as a supplement to the user’s scoring data.To solve the above problems,this thesis uses the attention mechanism to capture users’ attention to different objects,deeply studies the characteristics of users’ preferences changing with time,and puts forward relevant algorithms.The specific research contents are as follows:(1)Aiming at the homogenization of traditional social recommendation,this thesis proposes a preference attention model with the help of attention mechanism.The model calculates the user’s attention score,and then obtains the user’s attention weight to different friends,so as to obtain the influence of social factors on user preferences.At the same time,this thesis also sorts the items of user interaction according to the time series,and captures the user’s item preference through gated cyclic neural network.Finally,a recommendation algorithm based on user preference and social influence is proposed,which integrates users’ social influence preference and project preference,and makes Top-k recommendation to users through multi-layer perceptron.The experimental results on ciaodvds and douban data sets show that the proposed algorithm improves the hit rate and normalized loss cumulative gain compared with other comparison algorithms.(2)Aiming at the problem that social tags are not fully utilized,this thesis obtains the attention weight of users’ different tags of the project through the attention mechanism.At the same time,the intersection of user tags and item tags is used to enhance the characteristics of users and items.Considering the time-varying factors of user preferences,this thesis combines the time information of user interaction with the project and the characteristics of the project as the embedding matrix,and obtains the user’s time dynamic preferences through convolution neural network.Finally,this thesis proposes a recommendation algorithm based on time information and social tags,which integrates user preference features and project features into transfm for score prediction.The experimental results on Movielens-10 M and ml-latest-small data sets show that the root mean square error and average absolute error of this algorithm are reduced compared with other comparison algorithms.There are 25 figures,14 tables,and 81 references in this thesis.