[1J2-J-6] Web mining 1

[1J2-J-6-01] Recommendation System based on Generative Adversarial Network \\with Graph Convolutional Layers

Takato Sasagawa¹, 〇Shin Kawai¹, Hajime Nobuhara¹ (1. University of Tsukuba)

Keywords:recommendation system, Graph Convolution, Generative Adversarial Network (GAN)

A Graph Convolutional Generative Adversarial Network (GCGAN) is proposed to effectively recommend to new users or items. To maintain scalability, the discriminator is improved to capture latent features of users and items by using graph convolution from a minibatch size bipartite graph. Through the experiment using MovieLens dataset, it is confirmed the effectiveness of the proposed GCGAN compared with the conventional methods.