Tensor Factorization Recommendation Algorithm Based On Tag Information And User Trust

The widespread use of popular tagging websites has made the tag-based recommendation algorithm a hot topic in the research field of recommendation systems.The tag data contains a large amount of resource feature information,and also contains the user's preference information for the resource.Applying the tag data to the recommendation system helps to alleviate the data sparseness problem of the user item scoring matrix and improve the recommendation quality.The traditional tag-based recommendation algorithm cannot fully express the three-dimensional relationship among the three elements of user,item,and tag.Therefore,the tensor decomposition model is introduced into the tag recommendation algorithm.The existing tag recommendation algorithm based on tensor decomposition cannot achieve ideal recommendation effect due to the sparseness of tag data in the tag system.In this paper,the trust relationship between users and the tag recommendation algorithm based on the tensor decomposition method are combined,an improved tag recommendation algorithm based on tensor decomposition is proposed,and the tensor decomposition results are modified by using the user's trust similarity.The data matrix sparseness problem in the tag recommendation system is well alleviated.The main work of this paper is as follows:(1)Introduced the research background of this paper and the development situation of related topics at home and abroad.It also analyzed several mainstream recommendation algorithms in the recommendation system,highlighting the recommendation algorithm based on tags.(2)Introduced tag co-occurrence,Bisecting K-means clustering,tensor decomposition and user trust,etc.(3)Proposed a tag tensor decomposition model based on tag co-occurrence and Bisecting K-means clustering.The model uses the phenomenon of co-occurrence of tags to deeply mine the relationship between tags and form the feature vector of tags.The obtained tag feature vector is used as the input of the Bisecting K-means clustering algorithm to obtain clustered K tag clusters.The resulting tag cluster is composed of user data and resource data(users,resources,tag clusters)triplets,and high-order singular value decomposition is performed to obtain tensor decomposition data.The data matrix in the tag recommendation algorithm is often very sparse.This model effectively alleviates the data sparse problem.(4)Proposed a concept of trust similarity,and merged it with the results of tensor decomposition to obtain revised data.In this paper,user's rating data is used to calculate the global trust and local trust between users.The linear weighted method is used to combine the two as the overall trust degree of the user in the recommendation system,and the trust degree is combined with the user's interest similarity to form a The user trusts the similarity.Finally,the user trust similarity is merged with the tensor decomposition result to form the final recommendation result.(5)Use the data set in Movie Lens website to compare the proposed algorithm with other several tensor-based tag recommendation algorithms.The experimental results show that compared with the traditional tag recommendation algorithm based on tensor decomposition,the recommendation algorithm proposed in this paper which combines trust information,tag clustering and tensor decomposition algorithm has a better recommendation in the tag recommendation system.effect.