Research And Implementation Of Relation Extraction Method For Police Records Of Theft Cases

In real society,theft cases are frequent,which seriously endangers the property security of our citizens and the stability of social security.The current public security organs has obtained a large number of theft police records from various reporting channels,these records contain the location of the crime,the time of the crime,the stolen materials,crime characteristics,and other information,maximize the use of this information is one of the key factors for the police to quickly solve the case.However,theft police records are generally unstructured text records,and it is inefficient to extract important information from a large number of theft police records by manual means,and it is not easy to automate the system.If the relation extraction technology can be used to automatically extract entities and their relations with each other from police records,it will not only improve the efficiency of information extraction and enable the police to sort out the case more quickly,but also be important for the subsequent deeper use of this information.In this paper,we analyze some of the police record data of theft cases in a city,conduct an in-depth study to extract the key information,especially the relation information between entities,propose a method to extract entity relations applicable to police records of burglary cases and develop a case analysis system based on it.The specific points of work and innovation of this paper are as follows:（1）Designed and constructed a relational extraction dataset about theft casesAfter obtaining some of the police records data of theft cases provided by a city public security bureau,under the professional guidance of criminal investigation case officers,we first defined the common entity and entity pair relations of police records of theft cases.On this basis,we completed the relation triad annotation of 4010 police records of theft cases by manual annotation and performed corresponding data pre-processing to build the relation extraction dataset.（2）Designed and implemented a relation extraction modelA new joint entity-relation extraction model LWS is proposed,which consists of three main stages:first,the LERT model is used to generate word vectors of police records of burglary cases,and the generated word vectors are normalized using the weight normalization method.Then,we used the binary tagging method based on the Bi-SRU network to identify the head entities.The general process of this method is as follows:（1）Capturing recorded bidirectional semantic dependencies through Bi-SRU networks;（2）Two binary classifiers are used to predict the probability of each position in the record as the start and end position of the entity,and the binary tagging is completed according to the set threshold;（3）The entity identification task is completed by the proximity matching method.Finally,the previously predicted head entity information is incorporated into the word vector,and the recognition of tail entities is completed and again using the binary labeling method based on the Bi-SRU network under each specific relations,and finally,the relation extraction is completed for the police records of burglary cases.The results of the relation extraction experiments on the theft police record data set show that the proposed model has a 4.39%increase in recall and a 1.55%increase in F₁value compared to the baseline model,thus proving that the proposed model has a better extraction effect on the police theft records.（3）Designed and implemented a case analysis systemWe designed and developed a case analysis system for theft cases using frameworks such as Springboot and Vue.js.The system contains functions such as case relation extraction and similar case retrieval,which can provide some help to the police in their daily case processing,especially the analysis of concurrent cases.