CnGAN Generative Adversarial Networks for Cross-network user preference generation for non-overlapped users

Intro

cross-network Recommendation(are more robust against cold-start and data sparsity issues)

problem

However, despite the growing success of cross-network recommender solutions, the majority of existing solutions can only be applied to users that exist in multiple networks (overlapped users). The remaining non-overlapped users, which form the majority are unable to enjoy the benefits of cross-network solutions.

contribution

  • To the best of our knowledge, this is the first attempt to apply a GAN based model to generate missing source network preferences for non-overlapped users.
  • We propose CnGAN, a novel GAN based model which includes a novel content loss function and user-based pairwise loss function for the generator and recommender tasks.
  • We carry out extensive experiments to demonstrate the effectiveness of CnGAN to conduct recommendations for non-overlapped users and improve the overall quality of recommendations compared to state-of-the-art methods.

model