The wide application of artificial intelligence technology in the judicial field has brought profound changes to traditional legal practice. In this study, a public interest litigation evidence review and processing system based on Transformer and BERT model is constructed, and joint modeling of law recommendation and charge prediction is realized through a multi-task learning framework that fuses lawformer information. The methodology adopts Lawformer pre-training model for text encoding, combines the interactive attention mechanism to fuse the semantic information of the legal articles, and establishes the constraint relationship between the legal articles and the charges through the task-dependent constraint layer. The experimental results show that the MTL-LA-LJP model improves the accuracy of 0.130 in the law prediction task and 0.11 in the charge prediction task compared to CNN, and the performance advantage is more significant under the condition of small-sample data (1% training data), and the accuracy of the law prediction reaches 0.61. The study confirms the computer vision technology’s effectiveness in the review of public interest litigation evidence, and provides an opportunity for the construction of intelligent justice. The study confirms the effectiveness of computer vision technology in the review of public interest litigation evidence, and provides technical support for the construction of intelligent justice.
