Research On Sequence Recommendation Model Integrating User’s Long And Short-Term Interests

The recommendation system(RS)is suitable for scenarios where the user does not have a clear consumption goal or the needs are ambiguous.It can understand the user’s behavioral intentions or needs based on past user statistical characteristics,interaction habits,and historical behaviors,so as to help users make recommendations.Traditional RS based on collaborative filtering or content have problems such as relying on product label data and ignoring the order of interaction.Therefore,methods based on user interaction sequence modeling have become the mainstream implementation of recommendation systems.The main research question of this paper is based on the user-commodity interaction sequence data,the dependency between interaction items and the user’s long-term interest and short-term interest are obtained through the neural network model,and the two parts of interest are adaptively merged by dynamically adjusting the weight of the user’s long-term and shortterm interest.Generate the final recommendation list in descending order of weight,thereby improving the accuracy of model recommendation.After the analysis of the current recommender systems,this article includes the following two parts: user long-term and short-term interest capture and user long-term and short-term interest integration.For short-term interest capture tasks,a network model based on LSTM is proposed.Aiming at the problems of irregular time intervals and non-standard interaction semantics in interaction sequences,dynamic time-aware controllers and dynamic content-aware controllers are proposed as the optimization of LSTM networks.,So as to capture the shortterm interest of users;for the long-term interest capturing task,the interaction sequence is too long and the calculation complexity is too large.The sequence length and time threshold are divided into shorter calculation windows.For each window,the hierarchical attention model is used to obtain interest expressions with different granularities,and the multi-head selective attention model is used to weight the interest representations of all calculation windows.Capture the long-term interest of users.For the interest fusion task,a network model based on the self-attention mechanism is proposed for dynamic user interest fusion.The short-term interest candidate set and the long-term interest candidate set of the user are used as the input of the fusion model,and the two are obtained after training.The weight coefficient of,so as to dynamically adjust the ratio of two kinds of interests in different scenarios,and generate the final recommendation list.Finally,this article uses the public data set of Amazon e-commerce website to conduct experiments and compares with other existing models.The results verify the effectiveness of the model proposed in this article.