Deep Semantic Understanding and Accurate Classification of Public Interest Litigation Case Texts

With the burgeoning number of public interest litigation cases, their effective classification has become imperative. However, this task is fraught with challenges, given the inconsistency in text length, the intricate nature of legal terms, and the structural complexity within the text. To mitigate these hurdles, this paper presents a BERT-Att-MultiTextCNN-based text classification model designed for public interest litigation cases. The model capitalizes on BERT's deep semantic understanding capabilities, extracting key textual information via the Attention mechanism. Additionally, it employs the MultiTextCNN model to capture local features, thereby ensuring a more precise text classification. This model was validated using the datasets of public interest litigation cases, THUCNews, and SST-2, achieving accuracy levels of 93.5%, 94.8%, and 94.1%, respectively. The experimental findings underscore this model's significant advantages over conventional text classification models. The study's outcomes suggest that the proposed BERT-Att-MultiTextCNN-based model holds promising potential for text classification tasks, especially those involving the categorization of text in public interest litigation cases.