Developing Responsible and Reliable Entity and Relation Extraction Methods: A Human-in-the-Loop Perspective

Yang, Pingjing

Permalink

https://hdl.handle.net/2142/121576 Copy

Description

Title

Developing Responsible and Reliable Entity and Relation Extraction Methods: A Human-in-the-Loop Perspective

Author(s)

Yang, Pingjing

Issue Date

2023-12-14

Keyword(s)

• Relation Extraction
• Human-in-the-loop
• Responsible AI

Abstract

In this digital age, information extraction is still a fundamental task as the basis of more complex and user-oriented tasks, such as chatbots and search engines. Although efforts have been made in this area, adopting information extraction for a broad audience is still an underdeveloped problem. Both algorithm practitioners and end users still lack easy-to-access guidelines for using information extraction tools. This proposal aims to advance information extraction application research by developing generic and domain-specific (e.g., education science) information extraction tools to enable practitioners to apply relation extraction on their own corpus. This proposal starts with analyzing needs expressed by online social media users during an urgent social event, the Ukraine-Russia Conflict, using Twitter (X.com) data. Then, I will systematically evaluate state-of-the-art relation extraction approaches on public relation extraction datasets, including newswire, wiki, and web text. Finally, I will apply optimized models developed on public datasets to domain-specific text data in education science and involve education experts in the development life cycle. I expect to develop guidelines for practitioners to leverage better relation extraction tools on different types of corpus and insights on involving human experts in the loop of algorithm method development.

Type of Resource

text

Language

en

Owning Collections