Research On Graph Collaborative Filtering Model Combining Time Factor And Attention

The rapid development of Internet technology has led to the explosive growth of information.The massive amount of information meets the needs of users for information,but at the same time it also brings about serious information overload problem.As a powerful means to solve the problem of information overload,the recommendation system can provide users with information and items that they are interested in in a personalized way according to the user's individual needs,hobbies,historical behavior and other information.The early methods of recommendation algorithms are mostly based on collaborative filtering ideas and item content characteristics.Recently,with the success of deep learning,attention mechanism,and graph convolutional networks in their respective fields,researchers have begun to try to apply these emerging technologies into recommendation systems,and pioneered the design of many novel and effective recommendation models.At present,using new representation learning techniques or introducing auxiliary information to improve the representation ability of embedding has become the core content of the recommendation algorithm research.Graph structure data contains rich structure information and high-level relational information,and has attracted more and more attention from researchers in the field of recommendation systems.Based on this,Neural Graph Collaborative Filtering(NGCF)designed a recommendation model that recursively propagates embedding information on the graph structure,and captures collaborative signals by exploring the high-level connectivity between users and items.However,NGCF does not consider the influence of time context on user preferences in the process of embedding information propagation,nor does it distinguish the contribution of different neighbor node information to the target node.In order to solve these two problems,this paper proposes a graph collaborative filtering model combing time factor and attention TAGCF on the basis of NGCF.The model uses the time factor to integrate time information into the process of embedding information propagation,and uses the attention mechanism to distinguish the influence of embedding information from different neighbors.Specifically,the embedding of each first-order neighbor is multiplied by a time factor and an attention coefficient to form two kinds of embedding information.The time factor is calculated based on the time when the interaction occurs,and the attention coefficient is determined based on the similarity.Then the two kinds of embedding information are added together,and the embedding information propagated by all neighbor nodes is aggregated by the aggregation function to enhance the representation of the central node.Repeat this process to carry out high-level information propagation and aggregation,use the attention mechanism to combine the embedding representations obtained at different propagation levels,and finally generate the user's prediction rating for the items through the inner product interaction.Through comparative experiments with multiple baseline methods such as NGCF on the two recommendation system datasets Movie Lens and Amazon-books,Recall?Precision?Hit?and NDCG are used as evaluation metrics to verify the effectiveness of TAGCF and the effectiveness of time information and attention.