[1Win4] Poster session 1

[1Win4-59] Applying Independent Component Analysis for Business Risk Classification and Quantification

〇Akito Suzuki¹, Haruna Sugimoto², Yusuke Tashiro¹ (1.Mitsubishi UFJ Trust Investment Technology Institute Co., Ltd., 2.Doshisha University Graduate School of Science and Engineering)

Keywords:Independent Component Analysis, Business Risk, Securities Reports, Sentence Embedding

In this study, we aim to classify and quantify the ’business risks’ listed in the securities reports of various companies. Business risks are generally presented as free-form text and vary depending on the company’s business activities, meaning that they lack a clear classification and cannot be disclosed in quantitative data. To address this, we compute sentence embeddings of the business risk texts and extract independent components from the embeddings through ICA. This approach allows us to classify business risks and quantify their magnitude by analyzing the independent component scores for each company. Moreover, by utilizing higher-order correlations between independent components, we enhance the interpretability of the relationships between different business risks through network construction and clustering. As a result of the experiment, the independent components were qualitatively meaningful and convincing.