Abstract
Each company engages in Investor Relations (IR) activities by disclosing information about the company to institutional investors and individual investors through IR documents such as financial statements, securities reports, and CSR reports. Understanding external evaluations of the content disclosed by the company is crucial during IR activities. Especially, evaluation of financial analysts at the business segment level, is considered useful. Our research aims to develop a method for analyzing sentiment of financial reports at the business segment level. Here, how to extract sentences or pragprahs related to each business segment is an important issue. Existing approaches for such information extraction often rely on searching for business segment names. However, these approaches do not address the extraction of sentences that mention related companies or specific services and products. Therefore, we propose improving existing approaches by leveraging large language models (LLMs) and utilizing business segment descriptions within securities reports. Experimental evaluation demonstrates that our approach is useful to improve the sentiment analysis performance of financial reports at the business segment level.
