Research On Sequential Recommendation Model Based On User Long-term And Short-term Preference

With the development of Internet information technology,human beings have entered the era of information overload.By virtue of its advantages in personalized information screening,the recommendation system can help users make accurate decisions.To a certain extent,the problem that information providers and demanders cannot match accurately caused by information overload can be alleviated.In recent years,sequential recommendation based on user behavior has been widely concerned by academia and industry,and it also reflects some problems existing in current sequential recommendation.First,traditional recommendation models usually model user-item interaction behavior in a static way,and ignore the context information that has an impact on user preferences,so it cannot simulate the dynamic change of user preferences.Second,the existing sequential recommendation model only models the long-term preference or short-term interest of users,which is insufficient to explore the user preference.Third,the existing sequential recommendation models ignore the personalized behaviors of different users,which leads to the model cannot fully capture the interest drift problem caused by the dynamic change of user preferences.Therefore,this paper conducts research based on the sequential recommendation model of users’ long-term and short-term preferences.The main contents are as follows:(1)Aiming at the problem that the traditional recommendation model ignores the context information in user-item interaction and cannot simulate the dynamic change of user preferences,a long-term and short-term preference sequence recommendation model based on context information(CI-LSPSRM)is proposed.Firstly,according to the diversity of user interaction behaviors,different types of interaction behaviors reflect users’ different degree of preference for items.An embedded layer based on user interaction behaviors is designed to identify the similarities between user preference items.Secondly,the LSTM model is initialized according to the context information of user-item interaction behaviors,and the user interest factor is proposed depending on the context information of user interaction interval.Based on this,the LSTM model is improved to model the long-term and short-term preferences of users,and simulate the dynamic change of user preferences.Finally,in the recommendation prediction layer,trilinear product is used to integrate users’ long-term and short-term preferences and generate a list of top N recommendation candidates.Experiments were carried out on 4 subsets of Amazon’s public data set,and the AUC value and recall rate index were used for comprehensive evaluation.The experimental results show that the proposed model performs better than other advanced benchmark models and effectively improves the recommendation performance.(2)Aiming at the problem that the existing sequential recommendation models ignore the personalized behaviors of different users,which leads to the model cannot fully capture the interest drift caused by the dynamic preferences of users,a long-term and short-term preference sequential recommendation model based on user behavior(UB-LSPSRM)is proposed.Firstly,the dynamic category embedding of users is generated according to the category and time information of interactive items in the user’s sequence,so as to effectively establish the correlation between items and reduce the sparsity of data.Secondly,according to the time interval information between the user’s current click and the last click,the personalized timing position embedding matrix is generated to simulate the personalized aggregation phenomenon of users,so as to better reflect the dynamic change of user preferences.Then,the user’s long-term preference sequence integrated with the personalized temporal position embedding matrix is input into the gated cycle unit with session as unit to generate the user’s long-term preference representation,and the user’s long-term and short-term preferences are integrated through the attention mechanism to generate the user’s final preference representation,so as to achieve the purpose of fully capturing the user’s preferences.Finally,the user’s final preference is input into the recommendation prediction layer for recommendation prediction.Experiments were carried out on 7subsets of Amazon’s public data set,and the AUC value,recall rate and accuracy rate indexes were used for comprehensive evaluation.The experimental results show that the proposed model performs better than other advanced benchmark models and effectively improves the recommendation performance.