Temporal Knowledge Graph Construction from Text

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K17816
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Japan Advanced Institute of Science and Technology
• Principal Investigator: Kertkeidkachorn Natthawut 北陸先端科学技術大学院大学, 先端科学技術研究科, 講師 (40818100)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: Temporal Knowledge Graph / Financial KGs / Knowledge Representation / Relation Extraction

Outline of Final Research Achievements

The project outcomes include both the research methods and the development of temporal knowledge graphs in the financial domain. Specifically, we proposed a knowledge extraction framework that uses both semantic and syntactic features to extract useful knowledge from text, ensuring high-quality knowledge graphs. We constructed FinKG and FinKG-JP, temporal knowledge graphs in the financial domain. Temporal information, including report details and stock prices, was extracted using our financial ontology template and encoded at the edges of the knowledge graphs. We demonstrated the usefulness of FinKG in two applications: knowledge retrieval and stock price prediction.

Free Research Field

知能情報学関連

Academic Significance and Societal Importance of the Research Achievements

このプロジェクトでは、金融分野における時系列知識グラフであるFinKGとFinKG-JPの構築および開発方法を提案した。これらの時系列知識グラフは、金融分野における時間認識型AIアプリケーションの開発に貢献することができる。本研究の科学的意義は、時系列知識グラフの構築方法を進展させることであり、これは金融における予測分析や意思決定プロセスの向上に不可欠である。