Time Interval Aware Collaborative Sequential Recommendation With Self-Supervised Learning

With the continuous development of information technology and artificial intelligence technology,recommender systems have become an important component of many applications in the real world.In various types of recommender systems,sequential recommendation has received great attention.The general idea of sequential recommendation is to use the user’s historical interaction to predict the next item that the user may interact with.Among them,time interval information can reflect the user’s interest transfer process and has attracted the attention of many researchers.The accuracy of recommendation results is directly related to the success of an application,therefore,this task has received widespread attention from academic and industrial,and has important research significance and value.With the continuous development of deep learning technology,sequential recommendation has reached a relatively high level,but still faces challenges such as excessive attention to recent interaction items,data sparsity,and a lack of personalized user information fusion.In response to the above issues,this article conducts research on sequential recommendation problems based on time intervals.The main research content and innovations are as follows:(1)Time interval aware Attention score Adjustment method has been proposed in this paper.For the problem of overly focusing on recent interaction items,use time interval information combined with time kernel functions to simulate the impact of different user interactions on the final recommendation.For the problem of data sparsity and the lack of personalization in user information fusion,self supervised technology is used to achieve the fusion of sequence information and collaborative information,alleviating the problem of data sparsity and making user information fusion more personalized.(2)Time Interval Aware Collaborative Sequential Recommendation with Self-Supervised Learning has been proposed in this paper.For the problem of data sparsity in the process of user information generation,the user category interaction graph is constructed by introducing category information to alleviate the problem,and the user representation with category information and the user representation with item information are fused by using the self supervised learning technology to avoid the problem of over smoothing during information generation and obtain the final user representation of collaborative information.In response to the lack of personalization in user information fusion,a user information selection module is proposed,which completes the fusion of user information through a personalized gating network.This paper effectively extracts and integrates multiple aspects of user information,and combines time information to improve the user representation ultimately used for recommendation.This article conducted a large number of experiments on three datasets with different user numbers and duration distributions.The experimental results showed that compared with existing methods,the two models proposed in this article HR@10 and NDCG@10 The performance indicators are superior to existing methods.