BERT-Based Ensemble Model for Statute Law Retrieval and Legal Information Entailment

Hsuan Lei Shao, Yi Chia Chen, Sieh Chuen Huang*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

The Competition on legal information extraction/entailment (COLIEE) is an international information processing and retrieval competition. As an aid to future participants as well as question designers, this article describes how to connect legal questions taken from past Japanese bar exams to relevant statutes (articles of the Japanese Civil Code, Task 3) and how to construct a Yes/No question answering system for legal queries (Task 4) incorporating background materials on Japanese law. We restructured the given data to a dataset which contains all possible combinations of queries and articles as continuous strings as our samples. In this way, the difficult pairing task has been turned into a simpler classification task and samples for training became sufficient in number. Next, we used three BERT-based models to solve binary questions in order to achieve stable performance. As a result, the model achieved an F2-score of 0.6587 in Task 3 (ranked 1st) and an accuracy of 0.6161 in Task 4.

Original languageEnglish
Title of host publicationNew Frontiers in Artificial Intelligence - JSAI-isAI 2020 Workshops, JURISIN, LENLS 2020 Workshops, 2020, Revised Selected Papers
EditorsNaoaki Okazaki, Katsutoshi Yada, Ken Satoh, Koji Mineshima
PublisherSpringer Science and Business Media Deutschland GmbH
Pages226-239
Number of pages14
ISBN (Print)9783030799410
DOIs
Publication statusPublished - 2021
Event12th International Symposium on Artificial Intelligence supported by the Japanese Society for Artificial Intelligence, JSAI-isAI 2020, International Workshop on Logic and Engineering of Natural Language Semantics, LENLS 2020, 14th International Workshop on Juris-informatics, JURISIN 2020 - Virtual, Online
Duration: 2020 Nov 152020 Nov 17

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12758 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference12th International Symposium on Artificial Intelligence supported by the Japanese Society for Artificial Intelligence, JSAI-isAI 2020, International Workshop on Logic and Engineering of Natural Language Semantics, LENLS 2020, 14th International Workshop on Juris-informatics, JURISIN 2020
CityVirtual, Online
Period2020/11/152020/11/17

Keywords

  • BERT-based ensemble model
  • COLIEE 2020
  • Information retrieval
  • Legal AI
  • Legal analytics
  • Textual entailment

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'BERT-Based Ensemble Model for Statute Law Retrieval and Legal Information Entailment'. Together they form a unique fingerprint.

Cite this