Multi-task Learning for Automatic Event-Centric Temporal Knowledge Graph Construction

• Published in: Research Challenges in Information Science Vol. 446; pp. 811 - 818
• Main Author: Knez, Timotej
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Lecture Notes in Business Information Processing
• Subjects: Information retrieval; Knowledge graphs; NLP; Temporal relations
• ISBN: 3031057597; 9783031057595
• ISSN: 1865-1348; 1865-1356
• DOI: 10.1007/978-3-031-05760-1_59

An important aspect of understanding written language is recognising and understanding events described in a document. Each event is usually associated with a specific time or time period when it occurred. Humans naturally understand the time of each event based on our common sense and the relations between the events, expressed in the documents. In our work we will explore and implement a system for automated extraction of temporal relations between the events in a document as well as of additional attributes like date, time, duration etc. for placing the events in time. Our system will use the extracted information to build a graph representing the events seen in a document. We will also combine the temporal knowledge over multiple documents to build a global knowledge base that will serve as a collection of common sense about the temporal aspect of common events, allowing the system to use the gathered knowledge about the events to derive information not explicitly expressed in the document.