In the integration of artificial intelligence technology with the judicial field, due to issues such as the complexity of legal texts and the large number of disputed issues, the labor dispute resolution mechanism still needs to be optimized and improved. In the retrieval of labor dispute cases, this paper proposes a similar case matching model consisting of four modules: an embedding layer, an attention layer interaction, a pooling layer, and a prediction layer. This model enhances the interactivity between text representations by adding an attention mechanism to the expression model. Additionally, it introduces a pairwise comparison (PairWise) task to promote the model’s learning of relative ranking position information, designing a multi-task training method that combines sentence pair ranking. In identifying the focal points of labor disputes, this paper employs a classification network module based on LegalBert for category prediction, utilizes the Skip-Gram model to achieve vector representation of text, and uses a BiLSTM-Attention representation module to operate on sentence matrices. Through the prediction output layer, the focal points of disputes are identified, thereby establishing a dispute focal point identification model based on multi-feature fusion. Subsequently, a statistical analysis module for labor dispute-related legal data was constructed, which, together with the proposed case matching model and dispute focus identification model, constitutes the labor dispute resolution system mechanism. After 25 rounds of training, the proposed labor dispute resolution system achieved an accuracy rate of 90.69% with a loss of only 21.40%.
