[4Rin1] Interactive Session 2

[4Rin1-36] Business Confidence Prediction for Analyst Report using Convolutional Neural Networks

〇Shota Takayama¹, Seiichi Ozawa^1,2, Takehide Hirose³, Masaaki Iizuka³ (1. Graduate School of Engineering, Kobe University, 2. Center for Mathematical and Data Sciences, Kobe University, 3. Sumitomo Mitsui DS Asset Management Company, Limited)

Keywords:Text Mining, Business Confidence Prediction, Natural Language Processing, Convolutional Neural Networks, Deep Learning

To decide valuable companies to be invested, investment trust and fund management companies, which manage funds deposited from investors, have collected information on company’s budget status and plans. However, the number of visit reports are usually too large even for skilled fund managers to easily derive reliable business outlooks and investment decisions. In this research, to alleviate fund managers’ and analysts’ commitment for the investigation and analysis, we propose a machine learning system that can support them to make accurate predictions on business outlook from collected visit reports. We attempt to predict business confidence for specific companies and industries using CNN that is expected to have good readability and robustness for polarity perturbation. As a result, we obtain 81.4% in classification accuracy for analysts’ reports provided by the Sumitomo Mitsui DS Asset Management Company, Limited. It has 5.7% better accuracy than the best baseline model using Word2Vec and SVM.