Classification on Sentence Embeddings for Legal Assistance

TL;DR

This paper explores using BERT-based sentence embeddings combined with a linear classifier to semantically segment legal documents into predefined rhetorical roles, aiming to improve legal assistance efficiency.

Contribution

It introduces a method applying BERT embeddings and class weighting strategies for legal document segmentation, addressing the challenge of class imbalance.

Findings

Weighted classes improve segmentation performance.

Achieved an F1 score of 0.22 on the task.

Highlights the impact of class frequency on results.

Abstract

Legal proceedings take plenty of time and also cost a lot. The lawyers have to do a lot of work in order to identify the different sections of prior cases and statutes. The paper tries to solve the first tasks in AILA2021 (Artificial Intelligence for Legal Assistance) that will be held in FIRE2021 (Forum for Information Retrieval Evaluation). The task is to semantically segment the document into different assigned one of the 7 predefined labels or "rhetorical roles." The paper uses BERT to obtain the sentence embeddings from a sentence, and then a linear classifier is used to output the final prediction. The experiments show that when more weightage is assigned to the class with the highest frequency, the results are better than those when more weightage is given to the class with a lower frequency. In task 1, the team legalNLP obtained a F1 score of 0.22.