Extraction of Judgment Elements from Legal Instruments Using an Attention Mechanism-Based RCNN Fusion Model

In the field of jurisprudence, judgment element extraction has become a crucial aspect of legal judgment prediction research.The introduction of pre-trained language models has provided significant momentum for the advancement of Natural Language Processing (NLP) technologies, with the Bidirectional Encoder Representations from Transformers (BERT) model being particularly notable for its ability to enhance semantic understanding in unsupervised learning.A fusion model combining BERT and an attention mechanism-based Recurrent Convolutional Neural Network (RCNN) was utilized in this study for multi-label classification tasks, aiming to further extract contextual features from legal texts.The dataset used in this research was derived from the "China Legal Research Cup" judgment element extraction competition, which includes three types of cases (divorce, labor, and lending disputes), with each case type divided into 20 label categories.Four comparative experiments were conducted to investigate the optimization of the model by placing the attention mechanism at different positions.At the same time, previous models were learned and studied and their advantages were analyzed.The results obtained from replicating and optimizing those previous models demonstrate promising legal instrument classification performance.