Research And Application Of Evidence Association Network Construction Method For Judicial Document

With the rapid development of digital technologies such as artificial intelligence,big data,and cloud computing,various industries are accelerating their digital transformation,which also promotes a shift towards digitization and intelligence in the judicial system.In recent years,the exploration of intelligent judicial adjudication has gradually deepened,with the core of judicial activities being the excavation and analysis of evidence.Evidence serves as the crucial factual basis for judgments and typically exhibits coherence and correlation.As such,a thorough examination of the connections between pieces of evidence is essential for case progress.In this context,constructing an effective evidence linkage network becomes the key to enhancing the efficiency of judicial adjudication activities.The construction of an evidence linkage network first requires the extraction of evidence entities from judicial documents.However,evidence extraction faces challenges such as nested entities,imbalanced positive and negative as well as hard and easy samples,and a limited training dataset.Secondly,based on the extracted evidence entities,it is necessary to explore the correlations between evidence for different case types,while also considering that some evidence may be unreliable or uncertain in certain special circumstances and cannot be part of an evidence chain.In light of the problems encountered in constructing an evidence correlation network,this paper proposes a method for constructing an evidence linkage network for judicial documents.The research can be divided into the following three parts:(1)Implementing an evidence extraction model based on machine reading comprehension: Considering the characteristics of evidence in judicial documents,we construct an evidence extraction model using machine reading comprehension combined with data augmentation techniques.The machine reading comprehension-based evidence extraction model leverages external knowledge and an improved loss function to enhance the performance of nested evidence entity extraction.In conjunction with data augmentation techniques,the model utilizes unlabeled data to expand the training set,thereby bolstering the model’s feature learning capabilities.(2)Constructing an evidence correlation network based on the Apriori algorithm: Using the evidence entities extracted in(1),we build an evidence linkage network by employing the Apriori algorithm to mine the relationships between pieces of evidence.To enhance the model’s accuracy,we incorporate the Dempster-Shafer(D-S)evidence theory to mitigate the unreliability and uncertainty between pieces of evidence.(3)Design and implementation of evidence association network construction system for adjudication documents: Based on the application requirements of legal subjects for evidence intelligence system,the above two evidence intelligence models are integrated and the evidence association network construction system is implemented.The system includes the functional modules of evidence entity extraction and evidence association network construction.The machine reading comprehension model,utilizing external knowledge and an improved loss function,not only alleviates the issue of low performance in nested entity recognition during the evidence extraction process but also focuses on difficult-to-classify samples to enhance extraction performance.In combination with data augmentation techniques,this approach effectively strengthens the model’s feature learning capabilities.Building on this foundation,the Apriori algorithm and D-S evidence theory are employed to mine the correlations between pieces of evidence and address their unreliability,effectively constructing an evidence correlation network.Experimental results on a judicial document dataset show that the F1 score of the evidence extraction model has increased by 2.12%,while also improving the accuracy of the evidence linkage network construction.Therefore,the combination of this correlation mining algorithm and evidence extraction model provides a more efficient and accurate means of evidence analysis for judicial adjudication activities.