Research on Prediction of Legal Decisions Based on Graph Neural Network with Attention Mechanism

The prediction of legal decisions has become a growing area of interest due to the increasing need for efficiency, transparency, and fairness in judicial processes. Traditionally, legal decision-making relies heavily on human expertise, which can be influenced by subjective interpretations, lengthy document reviews, and varying legal precedents, making consistency a challenge. Existing automated approaches often struggle to capture complex legal semantics, contextual relationships, and long-range dependencies in legal texts. To address these challenges, this research proposes a Graph Neural Network (GNN) with attention mechanism-based model that can effectively focus on the most relevant sections of legal documents, enabling more accurate predictions of case outcomes. The model incorporates hierarchical attention to process both sentence-level and document-level contexts, thereby enhancing interpretability and precision. For experimentation, widely used datasets such as the CAIL2018, COLIEE 2020, and EUR-Lex dataset were utilized, covering various legal domains and case types. The attention-based approach identifies key facts, legal arguments, and precedent references within the documents, assigning higher weights to critical information that influences the final decision prediction. Experimental results demonstrate that the proposed method outperforms traditional machine learning and baseline deep learning models, achieving significant improvements in accuracy of 97.7 %. This research highlights that attention mechanisms can not only improve predictive performance but also provide transparency in the decision-making process, which is crucial in legal applications.