Aiming at the problem of data sparseness in recommendation algorithms, social data are usually introduced as auxiliary information for social recommendation. The traditional social recommendation algorithms ignore users' interest transfer, which makes the model unable to describe the dynamic characteristics of user interests, and the algorithms also ignore the dynamic characteristics of social influences, which causes the model to treat long before social behaviors and recent social behaviors equally. Aiming at these two problems, a social recommendation model named SLSRec with dynamic integration of social information was proposed. First, self-attention mechanism was used to construct a sequence model of user interaction items to implement the dynamic description of user interests. Then, an attention mechanism with forgetting with time was designed to model the short-term social interests, and an attention mechanism with collaborative characteristics was designed to model long-term social interests. Finally, the long-term and short-term social interests and the user's short-term interests were combined to obtain the user's final interests and generate the next recommendation. Normalized Discounted Cumulative Gain (NDCG) and Hit Rate (HR) indicators were used to compare and verify the proposed model, the sequence recommendation models (Self-Attention Sequence Recommendation (SASRec) model) and the social recommendation model (neural influence Diffusion Network for social recommendation (DiffNet) model) on the sparse dataset brightkite and the dense dataset Last.FM. Experimental results show that compared with DiffNet model, SLSRec model has the HR index increased by 8.5% on the sparse dataset; compared with SASRec model, SLSRec model has the NDCG index increased by 2.1% on the dense dataset, indicating that considering the dynamic characteristics of social information makes the recommendation results more accurate.

Aiming at the complexity of triple data's inner relation, a co-clustering recommendation model based on the PARAllel FACtorization (PARAFAC) decomposition was proposed. The PARAFAC was used for tensor decomposition to mine the relevant relations and potential topics between the entities of multidimensional data. Firstly, triple tensor data was clustered by using the PARAFAC decomposition algorithm. Secondly, three recommendation models for different schemes were proposed based on collaborative clustering algorithm, and compared for obtaining the optimal recommendation model through the experiment. Finally, the proposed co-clustering recommendation model was compared with Higher Order Singular Value Decomposition (HOSVD) model. Compared to the HOSVD tensor decomposition algorithm, the PARAFAC collaborative clustering algorithm increased the recall rate and precision by 9.8 percentage points and 3.7 percentage points on average on the last.fm data set, and increased the recall rate and precision by 11.6 percentage points and 3.9 percentage points on average on the delicious data set. The experimental results show that the proposed algorithm can effectively dig out tensor potential information and internal relations, and achieve recommendation with high accuracy and high recall rate.

To improve the accuracy of the recommendation system, a Location-Based Asymmetric Similarity for Collaborative Filtering (LBASCF) recommendation algorithm was proposed for the problem that traditional Collaborative Filtering (CF) recommendation algorithm does not consider the location information. Firstly, the cosine similarity and the Location-Based Asymmetric Similarity (LBAS) between users were calculated by the user-item rating matrix and the user's historical consumption location; secondly, a new user similarity measure was obtained by fusing the cosine similarity and location-based similarity. The blended similarity could reflect the user's preferences in both location and interest. Finally, based on the ratings of the user's nearest neighbors, new items were recommended to the user. The effectiveness of the algorithm was evaluated by using a dianping dataset and Foursquare dataset. The experimental results on the dianping dataset show that, compared with CF, the recall and precision of LBASCF were increased by 1.64% and 0.37% respectively; compared with the Location-Aware Recommender System (LARS), the recall and precision of LBASCF were increased by 1.53% and 0.35% respectively. The experimental results show that LBASCF can achieve better recommendation quality of the system based on the application of location-based services than CF and LARS.

The item recommendation precision of social tagging recommendation system was affected by sparse data matrix. A tensor factorization recommendation algorithm combined with social network and tag information was proposed, in consideration of that Singular Value Decomposition (SVD) had good processing properties to deal with sparse matrix, and that friends' information could reflect personal interests and hobbies. Firstly, Higher-Order Singular Value Decomposition (HOSVD) was used for latent semantic analysis and multi-dimensional reduction. The user-project-tag triple information could be analyzed by HOSVD, to get the relationships among them. Then, by combining the relationship of users and friends with the similarity between friends, the result of tensor factorization was modified and the third-order tensor model was set up to realize the item recommendation. Finally, the experiment was conducted on two real data sets. The experimental results show that the proposed algorithm can improve respectively recall and precision by 2.5% and 4%, compared with the HOSVD method. Therefore, it is further verified that the algorithm combining with the relation of friends can enhance the accuracy of recommendation. What's more, the tensor decomposition model is expanded to realize the user personalized recommendation.