Font Size: a A A

Research On Social Recommendation Algorithm Based On Matrix Factorization

Posted on:2020-01-22Degree:MasterType:Thesis
Country:ChinaCandidate:W W ZhangFull Text:PDF
GTID:2438330575959325Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Recommendation system research has always been a hotspot in data mining,providing users with accurate and personalized recommendation services.The algorithms of local neighborhood recommendation algorithm and global matrix factorization model have been widely studied.However,there are still studies that combine the complementary advantages of these two methods to improve the accuracy of the recommendation,and there are still problems of data sparseness and cold start.Therefore,in order to solve these problems,we have carried out the following research,the specific work of this paper is as follows:1.It mainly summarizes the background significance and research status of the research,and then expounds the related technologies in the field of recommendation systems,including recommendation algorithms based on neighborhood and model,and summarizes the evaluation indicators commonly used in the recommended field.Finally,the application of matrix factorization model in social recommendation algorithm is introduced in detail.2.Neighborhood-based collaborative filtering only uses the common score between users when performing similar user selection,and the matrix sparsity increases with the increase of the user amount,which leads to the problem of low recommendation accuracy.Therefore,this paper proposes a convolutional depth learning model based on label weights and neighbors.It introduces labels and refines label weights in social relationships.Items with the same label often have similar features,which makes it easier to cluster clusters with similar interests.Together,the cosine similarity is used to measure the correlation between them,and the user eigenvector and the formula for establishing the characteristic equation are given.The item features are processed using a convolutional neural network while processing project feature vectors.The item feature equation established by this method is more accurate,which greatly improves the scalability,computational efficiency and recommendation quality of the proposed algorithm.3.The social recommendation algorithm still uses the dual trust relationship to improve the accuracy,ignoring the influence of the user's own factors,and the recurrent neural network of the attention mechanism can model the long-term storage in the sequence data,so this article takes into account the friends.The strength of trust and the strength factor of itself and the model of attention mechanism.A social network recommendation algorithm based on topic attention and matrix factorization is proposed.Firstly,the degree of influence of self-intensity factor on social relations is introduced.Then,the attention neural network is used to model text information and extract text features.Finally,the predicted score is obtained by probability matrix factorizationand recommended.The experimental results show that the recommended performance is significantly improved.
Keywords/Search Tags:Recommendation system, Collaborative filtering, Matrix factorization, Social recommendation, Deep learning
PDF Full Text Request
Related items