Research And Design On The Construction Method Of Event Evolutionary Graph In The Field Of Civil Justice

Powerful tools for knowledge mining in a variety of sectors have been made possible by the continued development of big data and artificial intelligence technology.To increase judicial effectiveness and realize judicial intelligence,it is especially important for the judicial field,which has amassed a large number of documents,to use natural language processing technology to extract the rational knowledge from those documents and build the rational map in accordance with the logical relationships contained therein.The key study objective of information processing tasks is to learn how to use contemporary information technology to mine the complicated semantics and logical knowledge included in legal documents since they are long texts in Chinese.Because of the research on the creation of graphs for adjudication documents is still in its early stages,and there are currently comparatively few studies that are pertinent for logical knowledge mining in adjudication documents.In order to examine how to extract subject information from the files and documents associated with adjudication in civil cases,as well as how to create Event Evolutionary Graph in the context of civil justice,this thesis uses the civil justice area as a scenario.The main thesis work and innovation as shown below:(1)This study suggests a neural network model based on two-layer BERT+CRF to perform the event extraction task in order to address the issue of event extraction challenges caused by scattered event elements induced by long adjudication documents and volumes.The event extraction operation must take into account the dependence between trigger words and argument items due to the complicated event structure and dispersed event constituent,for this purpose,the model presented in this paper splits the event extraction task into two pipeline-based tasks: trigger word identification and argument element recognition.After employing BERT+CRF to identify the trigger word for the marker,the model fuse the position information of the trigger word and recognizing the argument element.According to the experimental findings,the model’s experimental F1 value for event extraction from magisterial documents can reach 87.18%.(2)This study develops a Bi-LSTM+ACNN+CRF model based on double-labeling for causality extraction of civil adjudication documents to solve the issue of layered causality structures in civil adjudication documents that make deep causality extraction challenging.The document-level causality extraction task’s text is longer and contains more events than the sentence-level causality extraction problem does,which makes document-level relationship extraction more challenging.The model divides the task of causality extraction into two sequence annotation tasks,performing the initial causality semantic recognition using the Bi-LSTM network sequence annotation method.The results of the annotation are then fed into the ACNN network,which extracts the first and last character positions of the causal event pairs and then identifies the deep causal relationships that are contained in them.The experimental findings demonstrate that,with an F1 value of up to 84.31%,the model can successfully extract the causal linkages from the judgment papers.