Efficient document classification techniques based on large-scale judicial documents are crucial to current judicial intelligent application, such as similar case pushing, legal document retrieval, judgment prediction and sentencing assistance. The general-domain-oriented document classification methods are lack of efficiency because they do not consider the complex structure and knowledge semantics of judicial documents. To solve this problem, a semantic-driven method was proposed to learn and classify judicial documents. Firstly, a domain knowledge model oriented to judicial domain was proposed and constructed to express the document-level semantics clearly. Then, domain knowledge was extracted from the judicial documents based on the model. Finally, the judicial documents were trained and classified by using Graph Long Short-Term Memory (Graph LSTM) model. The experimental results show that, the proposed method is superior to Long Short-Term Memory (LSTM) model, Multinomial Logistic Regression (MLR) and Support Vector Machine (SVM) in accuracy and recall.

With the deepening of intelligence construction of the national judicial organization, massive judicial documents accumulated through years of information technology application provide data analysis basis for developing judicial intelligent service. The quality and efficiency of case handling can be greatly improved through the analysis of the similarity of judicial documents, which realizes the push of similar cases to provide the judicial officials with intelligent assistant case handling decision support. Aiming at the low efficiency of most document classification approach for common domains in judicial document classification due to the lack of consideration of complex structure and knowledge semantics of specific judicial documents, an efficient judicial document classification approach based on knowledge block summarization and Word Mover's Distance (WMD) was proposed. Firstly, a domain ontology knowledge model was built for judicial documents. Secondly, based on domain ontology, the core knowledge block summarization of judicial documents was obtained by information extraction technology. Thirdly, WMD algorithm was used to calculate judicial document similarity based on knowledge block summary of judicial text. Finally, K-Nearest Neighbors ( KNN) algorithm was used to realize judicial document classification. With the documents of two typical crimes used as experimental data, the experimental results show that the proposed approach greatly improves the accuracy of judicial document classification by 5.5 and 9.9 percentage points respectively with the speed of 52.4 and 89.1 times respectively compared to traditional WMD similarity computation algorithm.