Generating query-relevant business descriptions leveraging underrecognized businesses

Abstract

In corporate analysis, information tends to concentrate on core businesses that account for a large share of sales, while underrecognized businesses — those indicating diversification or emerging capabilities yet receiving little attention — are often buried and overlooked. We propose a method that classifies business sentences extracted from Japan’s annual securities reports into core businesses and underrecognized businesses, assigns segment-level sales weights, and leverages a large language model (LLM) to generate query-relevant business descriptions. Our pipeline first produces a business overview covering all of a company’s operations, including underrecognized ones, and then generates a description tailored to a user’s query that explains the relevance between the query and the company’s businesses. Evaluation with LLM-as-a-Judge confirmed that our method, which structures input through business classification and segment-level sales weights and incorporates the business overview, generates more detailed query-relevant business descriptions than a baseline that processes business sentences without classification.