| With the rapid development of the Internet,the amount of information has shown an explosive growth,and the information overload has greatly increased the difficulty for users to obtain information.By analyzing the user’s interaction history,the recommender system can capture the user’s preference more accurately,and then recommend items to the user to meet personalized needs.Users’ interactions with different items usually contain sequential order:previously interacted items often have an impact on the later.So the sequential pattern of user sequence should be taken into consideration.Sequential recommender systems view a user’s historical interactions as a sequence in which the items in the sequence are arranged in chronological order.By modeling successive items in the sequence,the item transition pattern can be captured and the next item that the user is likely to interact with in the future can then be recommended.The rise of deep learning has led to a trend of diversification of modeling approaches in the field of sequential recommendation.In recent years,in addition to sequence modeling algorithms such as Recurrent Neural Networks(RNN)and Transformer,Graph Neural Networks(GNN)have been increasingly applied to sequence recommendation work,but most of the current methods only target pairs of items within the current user sequence However,most of the current approaches only model the conversion patterns of items within the current user sequence.However,there are often some connections and interactions between users(items)in real life,i.e.,current user preferences may be influenced by other users(friends);similarly,item characteristics may be influenced by one or more items in other user sequences).In this paper,we propose a deep neural network sequence recommendation algorithm based on co-occurrence and social relationship,considering co-occurrence relationship of items between sequences,user social relationship and sequence context information.The main work includes the following three aspects:1)To address the problem that existing algorithms fail to consider item cooccurrence relation across different sequences,this paper proposed a sequential recommendation algorithm based on co-occurrence graph attention networks.First,a global graph of item co-occurrence relationships among sequences is constructed;then,the sequence items are modeled using graph attention neural networks to obtain higher-order features incorporated into the co-occurrence relationships,and the next possible items of interest to the target users are realized through the prediction layer.The performance of the proposed recommendation algorithm model is evaluated on two publicly available datasets,MovieLens-1M and Amazon Video Games,and the experimental results show that the proposed approach improves in all metrics compared to the baseline approach.2)To address the problem that existing algorithms fail to consider social relation and context information,this paper proposed a sequential recommendation algorithm based on hyper graph neural networks fused with social relations.Starting with multiple granularities,this approach introduces the effect of short-term user preferences consisting of the last few items on the modeling of individual items(i.e.a many-to-many relationship between items)on the basis of considering the pairing relationship between items in the same sequence(i.e.a one-to-one relationship).Besides,relationship between different users also has an impact on the modeling of item.Experiments on public datasets show that this method is superior to the baseline method.3)On the basis of the above theoretical research works,this paper builds a movie recommendation prototype system.This system recommends movies for the user by analyzing the historical interaction data of the currently logged-in user and combining the co-occurrence relationship and social information between items.. |
