Legal Case Retrieval by Essential Element Extraction Based on Reading Comprehension Model

Chen-Hua Huang, Chuan-Hsin Wang, Yao-Chung Fan, Fang-Yie Leu

Abstract: In this study, we propose a Legal Document Retrieval Pipeline. Given a legal case, we construct a scenario retrieval process based on various types of Essential Elements for Prosecution (EEP) associated with different criminal charges. We employ a reading comprehension model to extract essential scenario details, meeting the requirements of individual criminal charges. Subsequently, we extract keywords from these essential scenarios and utilize the embeddings of these keywords to compute the cosine similarity between each essential element, thus identifying the most closely related judgment documents. This approach dissects the overall direction of judgments into smaller components and derives similar judgments by matching the details within the judgment documents. In this study, we use the crimes of forgery and breach of trust as preliminary case types. We incorporate ChatGPT to assess the similarity between two judicial documents. We demonstrate that ChatGPT’s similarity judgments closely align with those of legal experts. The experiment results demonstrate the effectiveness of our legal document retrieval pipeline.

External IDs:dblp:conf/imis/HuangWFL24