Incorporate Transformers with Generative Adversarial Networks for User-based and Item-based Collaborative filtering recommendation

Abstract

GAN Based Heterogeneous Bibliographic Network Representation for Personalized Citation Recommendation (GAN-HBNR) is an existing GAN recommendation approach. However, GAN-HBNR is only focused typically on information from the target user or target item for creating user/item representation to predict the score, while ignoring the opinions of other users or neighbors. In common sense, the user’s neighbors and item’s neighbors make a significant effect on the user’s characteristics and item’s characteristics. To focus on the user’s neighbors and the item’s neighbors, I apply user-based collaborative filtering (user-based CF) and item-based collaborative filtering (item-based CF) to my proposed model. Furthermore, I use a transformer instead of denoising autoencoder (DAE) as the discriminator on GAN. Because DAE weighs every item that the target user has rated with the same attention, while transformer weighs every item that the target user has rated with the different attention. Therefore, my proposed model incorporates transformers in a generative adversarial network-based model to learn user representation and item representation that represents the relation between the target user’s preference and his/her neighbors’ preference to perform user-based CF, and the relation between the target item’s preference and its neighbors’ preference to perform item-based CF. The user representation and the item representation are used to predict rating score. To evaluate the proposed method, GAN-HBNR on movieLen Small Latest datasets is compared with the proposed method to achieve higher accuracy.