Research On Recommendation Algorithm Based On Convolutional Generative Adversarial Network

The rapid development of the internet brings people more convenience and choices.On the other hand,people are facing the problem of information overload.The research of Recommender Systems(RSs)has greatly eased the problem of information overload and played an important role in people's production and life.Generative Adversarial Networks(GANs)have recently been introduced into the domain of recommendation due to its ability of learning the distribution of users' preferences.But,existing GANs-based methods only exploit the user-item interactions,while ignore to leverage the information among the interacted items.Meanwhile,Convolutional Neural Network(CNN)has shown its power in learning high-order correlations,which can provide help for learning higher-order features among interactive items.In this paper,combining with the strengths of both GANs and CNN,we propose a Dilated Convolutional Generative Adversarial Network(Di CGAN)for recommender systems using vector-wise adversarial training.For generator,we first embed the interacted items of per user into an image in a latent space,then use several dilated convolutional filters and a vertical convolutional filter to capture the high-order correlations of items.Moreover,an attention module is employed before convolution to generate attention maps for adaptive feature refinement.Finally,we map the cascading features to user interaction vectors.Discriminator distinguishes the real user interaction vectors and generated interaction vectors,exploiting Multi-layer Perceptron(MLP)structure.Experiments on several public datasets verify the effectiveness of Di CGAN over stateof-the-art models in terms of top-N recommendation.Additionally,experiments also test the influence of CNN and verify that CNN can improve recommendation efficiency by extracting the characteristics of complex higher-order relations.