Knowledge-enhanced Document Representation Learning for Legal Judgment Support

The application of Artificial Intelligence (AI) to the legal domain, or LegalAI, has long been recognized as a potent tool for enhancing the efficiency of legal professionals. Despite this potential, prior research applied general AI models and neglected the knowledge in law. Such oversights have led to outcomes that were sometimes unexpected, including models exhibiting low performance and even contravening legal principles. This thesis addresses these challenges by focusing on the enhancement of legal knowledge representation within AI algorithms, aimed at supporting judicial work. This research explores three critical aspects: (1) Classification: The initial step in legal proceedings involves the categorization of submitted documents, a process characterized by its repetitive and significant time requirements. Previous studies overlook the detailed classifications within the legal domain in favour of broader categories. This research considers the detailed classification of cases, aiming to reduce the legal expert workload through AI. (2) Summarization: The ability to efficiently locate and summarize relevant prior cases is crucial for judges and lawyers. Existing AI models fail to leverage the specific summary structure and elements essential for creating precise and compact judgment summaries. This research proposes a method of structured summarization to respond to legal experts’ needs. Moreover, this approach can further improve the quality of the summary of other domains. (3) Prediction: Particularly in criminal law, AI models have the potential to identify anomalies in sentencing, such as unusually lenient or harsh prison terms. Prior approaches either overlook the legal knowledge or use it in inefficient ways. This research introduces a graph-based representation of statutory provisions to refine the model’s understanding of legal contexts, aiming for more accurate predictions of sentencing outcomes to find anomalous judgments. By addressing these areas, this thesis contributes to the development of LegalAI that is both effective and respectful of the domain’s expectations and principles.

Name thesis:degree_name

PhD

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Computer Science

Grantor dc:publisher

ResearchSpace@Auckland

Year dc:date.issued

2024

Author dc:creator

• Wang, Qiqi

Advisors dc:contributor.advisor

• Zhao, Kaiqi
• Amor, Robert
• Liu, Benjamin

Statement dc:rights

• Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Licence dc:rights.uri

• https://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm

Handle dc:identifier.uri

https://hdl.handle.net/2292/70811

OAI identifier oai:identifier

oai:researchspace.auckland.ac.nz:2292/70811