Research On Sequential Recommendation Algorithm For Perceiving On Long-Short Term Preferences Of Users

Sequential recommendations can provide users with accurate and personalized recommendations by modeling the sequence of user-item interactions.It can alleviate the problem of information overload,save time and energy for users,and create great value for enterprises or individuals.At present,many sequential recommendation models separate user behavior preferences into long-term behavior and short-term behavior.Although remarkable achievements have been achieved in previous studies,there are still the following deficiencies:(1)Researchers only focus on the sequential dependence of user interaction behavior,and ignore the impact of time context information on the next interaction.(2)In real application scenarios,the sequential recommendation model has been affected by the problem of data sparsity and sequential noise for a long time.Using a large number of unlabeled data for training will result in low quality of user representation.In order to alleviate the impact of these problems,this paper will study the integration of time and context information in sequential recommendation,alleviate sequential noise and enhance user representation.The main contents of this paper are as follows:(1)In order to capture the characteristics of time information in user interaction sequence and improve the effect of sequential recommendation,this paper proposes a sequential recommendation model(TCLSRec)that combines time context information with long-short term preference.The model uses the self-attention mechanism of perceiving time interval to capture the deep correlation between user’s short-term behavior sequence with time interval information.Then model uses the RT-GRU network with time interval gate to model user’s long-term behavior preference.The model designs a gating mechanism to dynamically allocate the influence weight of users’ long-term preferences and short-term preferences,and integrates the long-term and short-term behavior preferences.The experimental results on three real data sets show that the TCLSRec model in this paper solves the problem of timeliness after integrating the time interval information,and improves the recommendation performance.(2)In order to alleviate the problems of sparse data,unmarked data and large amount of noise in the sequence,this paper proposes an improved model(ILSRec)based on the TCLSRec model.The model models the user’s original interaction behavior sequence in the same way,and simultaneously designs a comparative learning task.This task generates the data enhancement sequence from the user’s original interaction sequence,and trains the enhancement sequence by using the method of comparative learning after modeling.So the model can obtain the self-monitoring signal of the original data for fitting.Finally,the model uses a multi-task training strategy to optimize the loss functions with the two training tasks together in a linear summation way,thus it can obtain higher quality user representation.The experimental results on four real data sets show that ILSRec model proposed in this paper can obtain a more accurate representation of users’ real preferences by training the self-supervised signals extracted from the sequence data,and further improve the accuracy and robustness of the recommendation result.Figure [29] Table [8] Reference [78]...