[1I3-GS-2] Machine learning: Market analysis

[1I3-GS-2-01] Extraction of Latent Event Topic Using Population Data and Topic Mode

〇Tomohiro Mimura¹, Shin Ishiguro^1,2, Satoshi Kawasaki^1,2, Daichi Shimizu¹, Yousuke Fukazawa¹ (1. NTT DOCOMO, INC., 2. Univ. of Tokyo)

Keywords:Topic model, Generative model, Population data

Predicting the number of visitors to an event is an important issue in reducing congestion. In this paper, we proposed a method to predicting the number of visitors. To predict visitor, we used neural topic model based on Joint Multimodal Variational AutoEncoder (JMVAE) and Student-t Variational Autoencoder with Implicit Optimal Priors. In the experiment, our proposed method showed higher prediction accuracy in root absolute error compared to conventional methods.