Financial Text Mining: Market Sentiment Analysis and Document Semantic Similarity for Different Languages

2022 Fiscal Year Final Research Report

• Project/Area Number: 18K11558
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 62020:Web informatics and service informatics-related
• Research Institution: Konan University
• Principal Investigator: Seki Kazuhiro 甲南大学, 知能情報学部, 教授 (30444566)
• Project Period (FY): 2018-04-01 – 2023-03-31
• Keywords: 深層学習 / 機械学習 / 大規模言語モデル / 景況感指数 / 足元予測 / 多言語モデル / 文書間類似度

Outline of Final Research Achievements

In this research, we studied (1) business sentiment forecast and (2) estimation of similarity between documents written in different languages, with the aim of promoting research on text mining in the financial and economic fields by utilizing a large amount of textual information such as news articles. For the former, we employed a model based on a self-attention mechanism, which has become the mainstream in recent natural language processing tasks, and achieved robust and highly accurate business sentiment prediction. For the latter, we successfully estimated the semantic document similarity between Japanese and English, and between English and Hindi, by using the internal representation of a translation model based on deep learning.

Free Research Field

人工知能

Academic Significance and Societal Importance of the Research Achievements

従来はアンケートなどのコスト・時間がかかる集計調査によっていた景況感が低コストかつほぼリアルタイムで行えることが示されたことにより，本研究成果を利用することで，金融当局の政策や企業の意思決定がよりタイムリーかつ効果的に行えるものと期待できる．さらに，異言語の文間の類似度を高精度で推定できることが明らかになったことから，これを発展させ前者と併用することで，言語の壁を超えて金融・経済関連テキストデータを統一的に分析することが可能となる．