Recommender systems suffer from interaction data sparsity in reality. Recently, generative adversarial network-based recommender systems have shown the potential to solve the problem. The negative sampling methods use the generator to collect signals from unlabeled data, while they suffer from sparse rewards in the policy gradient training process. The vector reconstruction methods generate user-related vectors for data augmentation to enhance robustness, but they lead to redundant calculation and ignore information conveyed by items. To alleviate the limitations of these methods, we propose a novel framework termed Personalized Recommendation with Conditional Generative Adversarial Networks to consider both of the user and the item subset as conditions. The sparsity and the dimension of conditional rating vectors can be controlled in our method, which simplifies both the generator’s reconstruction task and the discriminator’s learning task. In addition, the proposed method formulates conditional rating vector generation as a user-item matching problem, which allows a more flexible model selection for the generator. Experiments are conducted on three datasets to evaluate the effectiveness of the proposed framework.
展开▼
