[4E2-GS-2] Machine learning

[4E2-GS-2-03] Proposal of Product Image Generation Model based on CVAE Learning Abstract Information of Text Data

〇Ayuno Fuchi¹, Masaki Masuda¹, Masakazu Asano¹, Ayako Yamagiwa¹, Masayuki Goto¹ (1. Waseda University)

Keywords:Image generation, Product analysis, Latent Dirichlet Allocation, Conditional VAE

Each product on the EC site has attributes, image, and description and especially visual information has a large impact on customers' purchase decisions. Therefore, several studies to generate product images have been done recently. However, the abstract needs of customers cannot be reproduced with the generated images based on alone the specific attributes. On the other hand, some words included in the descriptions may express the customers' abstract needs. Therefore, if we can model the impact of abstract information expressed in descriptions on images and generate product images based on abstract information, we can capture customers' needs better. In this study, we propose an image generation method that combines Latent Dirichlet Allocation and Conditional Variational Autoencoder. We show the results of applying the proposed method to the real data and point out that it enables to generate product images based on abstract information extracted from product descriptions.