[3E3-GS-2] Machine learning: market analysis

[3E3-GS-2-03] Estimation of Latent Competition from Discrete Event Time Series

〇Shuichi Miyazawa^1,2, Daichi Mochihashi³ (1. The Graduate University for Advanced Studies, 2. AGC Inc., 3. The Institute of Statistical Mathematics)

Keywords:Bayesian Inference, Event Data, Differential Equations

To investigate the industrial market, we analyze patent information to estimate the latent competitive relationships between industrial firms. In existing studies, text analysis methods are often used for patent analysis to detect correlations between technologies and companies. However, existing methods cannot capture the competitive relationships between companies that might provide deeper insights for technology marketing. To solve this problem, we propose a model named Ordinary Differential Equation integrated Log Gaussian Cox Process (ODE-LGCP) that combines the ODE systems, for modeling competition in this study, with generative processes of event data. In the experiment, we use the real patent data and show that our proposed model can estimate competitive relationships within companies.