Applying BBLT Incorporating Specific Domain Topic Summary Generation Algorithm to the Classification of Chinese Legal Cases.

In response to the challenge that most existing case retrieval platforms can not effectively extract feature information of Chinese legal cases, and thus perform unsatisfactorily in terms of indicators such as relevance and accuracy of retrieval results. We propose to apply LBBT model incorporating domain-specific topic-based text summary generation algorithm to the classification of Chinese legal cases. In our proposed LBBT model, we use LDA to extract subject keywords for each type of legal documents separately, and then the TextRank algorithm is introduced to generate abstract for each legal document by combining the extracted subject words. BERT is used to vectorize the generated abstracts adopted as the inputs of BiLSTM to implement the task of classification on Chinese legal documents. The experimental result on the data set of 2500 single charge Chinese legal judgment documents obtained from CAIL2022 shows that our proposed LBBT model can effectively remove the redundant information in legal documents and improve the ability of LSTM to grasp the global key semantic information of long texts.