Separation And Extraction Of Complex Crime Events Based On Dependency Syntax And Semantic Model

Judicial cases are mainly composed of multiple basic criminal incidents that are related and combined.An important task in the judicial case handling process is to analyze the incidents of the case,clarify the basic criminal incidents of the defendant,and analyze,adjudicate,and send sentences for subsequent cases.Provide support,and can also effectively support the development and research of intelligent auxiliary case-handling systems.Therefore,in the judicial field,the extraction of criminal events of the defendant in a case is the main task of judicial intelligence.At the same time,event extraction is also one of the most concerned and benchmark tasks in natural language processing.Traditional event extraction pays more attention to the correctness of trigger words and argument types,and event extraction tasks have achieved good results in other fields at the sentence level.In the judicial field,the object and action of the event are the focus of attention.The task of extracting complex multi-person or multi-criminal crimes in the judicial field has the following characteristics: 1)Complex multi-person and multi-criminal cases involve multiple related persons,multiple characters usually participate in multiple criminal events,and there are a large number of event nesting and component sharing.2)Judicial documents have a certain structure and expressiveness.In order to express smoothness,some event descriptions omit subjects and objects.In view of the above characteristics,The specific research in this paper is as follows:(1)A joint model of trigger word recognition and classification based on semantics.At present,the Chinese event extraction system generally has the problem that the quality of the annotated event corpus is not high,and the ratio of false trigger words to true trigger words is too high.Therefore,this paper uses the maximum entropy model to combine the rules,vocabulary,syntax and semantic features of the indictment,and adds conditional random fields as constraints to construct trigger words in the judicial field.Recognition models designed to accurately identify criminal occurrences and types of crimes that occur in text.(2)Introduce dependency syntax to solve the problem of event omission.In the multi-person criminal prosecution documents in the judicial field,a large number of events are nested and components are shared,which can easily lead to the absence of event components and the omission of events.In this paper,dependency syntax analysis is introduced.Based on the trigger lexicon and dependency syntax tree corresponding to various types of criminal events,the verb grammar and syntactic structure are analyzed,so that each trigger event word corresponds to an event,and each event corresponds to a defendant.The problem of omission of events in the criminal indictment.(3)Analyze the default structure and formulate complementary rules to solve the problem of incomplete extraction of criminal events.Omission is a common phenomenon that actually exists in the Chinese language.Judicial documents with multiple persons or multiple crimes are more structured and expressive.Therefore,for the sake of clarity,some event descriptions have event components omitted.By constructing the SSDP(Syntactic Semantic Dependency Parsing)diagram,this paper summarizes several common default structures in multi-person or multi-criminal prosecution documents,and designs the corresponding rule complements to solve the lack of extraction components in multi-person or multi-criminal crimes.The problem.(4)The identification and extraction method design of special event element arguments.When analyzing,adjudicating,and sentencing a criminal case,some event elements often greatly affect the outcome of the case,such as the amount of theft in the crime of theft,and the number of grams of drugs involved in drug-related cases.In complex criminal cases,there are relationships between special arguments,such as intersection,belonging,and sharing.In order to accurately extract these event arguments and accurately assign them to the corresponding incident persons,for the special event arguments in different cases,In this thesis,relevant extraction algorithms and extraction rules are designed to assist extraction.In this paper,a method for separating and extracting complex crime events based on the dependency syntax and semantic model is proposed,which is verified by experiments and prototype systems respectively.The experiments show that the proposed method is better than the comparison method.The facts of the crimes corresponding to each defendant are extracted from the text of the multi-person or multi-crime indictment,which not only reduces the workload of the staff in the relevant fields,but also provides technical support for the precise sentencing of the intelligent auxiliary case-handling system.With advanced natural language processing technology for judicial efficiency and judicial justice.