A Novel Ensemble Approach for Legal Case Recommendation: Leveraging Sentence-BERT and Weighted Random Forests

Recommending legal case outcomes is a challenging task that demands advanced methods to decipher the intricate semantic nuances in legal texts. This paper introduces a novel approach that integrates Sentence-BERT (SBERT) embeddings with a weighted Random Forest Classifier, augmented with a cosine similarity-based feature to recommend case outcomes. To address imbalances in classes, the Synthetic Minority Oversampling Technique (SMOTE) is implemented, resulting in more consistent results. Experimental evaluations reveal that the framework attains exceptional performance in legal case classification, achieving an F1-score of 0.9537, Cohen’s Kappa of 0.9446, and an AUC-ROC of 0.9723. Accurate forecasting of case outcomes improves decision-making in legal analytics and supports the recommendation of pertinent legal precedents and strategic insights. The importance of semantic similarity and the possibility of hybrid AI approaches in improving legal decision support systems are typically highlighted by these findings.