Research On Decoupled Representation Method Of User Behavior For Sequential Recommendatio

Sequential recommendation algorithm recommends the next possible item of interest to the user according to the user’s past interaction sequence.One of its key technologies is to learn the implicit representation of sequence interactions to capture user preferences and provide users with accurate recommendation results.Based on the modeling of user-item interaction sequence,this paper uses the excellent scalability of a variational autoencoder and combines various deep learning methods to study how to improve the quality of user hidden variables to improve the recommendation performance of the model.The main work is as follows:(1)Recommendation models typically use users’ historical behaviors to obtain user preference representations for recommendations.Most of the methods of learning user representations always entangle different preference factors,while the disentangled learning method can be used to decompose user behavior characteristics.In this paper,a variational autoencoder-based framework DSVAECF is proposed to disentangle the static and dynamic factors from the user’s historical behaviors.Firstly,the two encoders of the model use multi-layer perceptron and recurrent neural networks respectively to model the user behavior history,to obtain the static and dynamic preference representation of the user.Then,the concatenate static and dynamic preference representations are treated as disentangled representation input decoders to capture user decisions and reconstruct user behavior.On the one hand,in the model training phase,DSVAECF learns model parameters by maximizing the mutual information between reconstructed user behaviors and actual user behaviors.On the other hand,DSVAECF minimizes the difference between the disentangled representations and their prior distribution to retain the generation ability of the model.The experimental results on Amazon and Movie Lens data sets show that compared with the baselines,DSVAECF significantly improves the normalized discounted cumulative gain,recall,and precision,and has better recommendation performance.(2)Most of the sequential recommendation models based on variational autoencoders use KL divergence as a regularization term to make the posterior distribution of latent variables of each user equal to the corresponding distribution.Such strong constraints make latent variables lose personalized information in the learning process.Therefore,we propose the DCWVAE model.Firstly,user preferences are decomposed into static and dynamic parts in the latent space according to the work in(1).To provide a more personalized recommendation,the maximum mean difference regularization method is introduced to constrain the posterior distribution of aggregated user latent variables instead of a single one.Then,we conduct contrastive learning as the auxiliary task to make the disentangled latent variables capture more personalized and significant features.Finally,to enhance the recommendation performance,the dropout-based regularization is also used to train the model.The extensive experimental results on three real-world datasets show that DCWVAE can achieve better recommendation performance and robustness against noisy data.