Joint Extraction Methods for Semantic Retrieval in Chinese Judicial Cases

The Supreme People’s Court of China has implemented the compulsory retrieval system for similar cases, making similar case retrieval the focus of research. Over 110 million judgment documents have been published on China Judgements Online, which provides massive data support. Recent approaches for semantic search are to embed queries and documents into semantic space and compute the similarity between them for ranking. These methods only represent the overall meaning of queries and documents, but learn a lot of redundant information from long text. In order to capture the case facts precisely from judgment documents, we propose a joint extraction model to build knowledge graphs for judicial cases. Experiments on recent judgment documents demonstrated the effectiveness of this knowledge graph construction process. The proposed case retrieval method can improve the efficiency for judicial trials, promote judicial justice, prevent the abuse of discretion, and unify the scale of adjudication.