ColdGAN: an effective cold-start recommendation system for new users based on generative adversarial networks
Journal
Applied Intelligence
Journal Volume
53
Journal Issue
7
Pages
8302-8317
Date Issued
2023-04
Author(s)
Abstract
Research on the problem of new user cold-start recommendation generally leverages user side information to suggest items to new users. This approach, however, is impractical due to privacy concerns. In this paper, we propose ColdGAN, an end-to-end GAN-based recommendation system that makes no use of side information to resolve the new user cold-start recommendation problem. The proposed ColdGAN explores the merit of GAN that enables precise data generation given imprecise data. Our generative network learns to predict item ratings that cold-start users would make in the future given their limited rating behavior data. The predicted ratings are evaluated by the discriminative network trained for determining whether the ratings are precise enough. Moreover, a novel rejuvenation function and relevant item loss are incorporated into ColdGAN to enhance the predictions made by the learned generative network. Experiments based on three real-world datasets demonstrate that ColdGAN significantly outperforms many state-of-the-art recommendation systems. Also, our designed rejuvenation function and relevant item loss are effective in guiding our generative network to infer item ratings of cold-start new users.
Subjects
Generative adversarial networks | New user cold-start problem | Recommendation systems
Publisher
SPRINGER
Type
journal article