Research On Element Extraction And Knowledge Expression For Criminal Case Record

Criminal case records are important carriers of text information of public security departments,the high-quality utilization of record can support the pre-prevention and precise strike against illegal and criminal act effectively.Telecom fraud has been on the rise in recent years and is an important part of criminal investigation cases,as a noncontact crime,the huge amount of records are the focus of investigation and handling of cases.However,the elements in the records are scatter throughout the text and mostly spoken,which makes the manual reading analysis time-consuming and energyconsuming,and limits the value of the records greatly.To solve above problems: firstly,this paper improves the accuracy of dependency parsing model on criminal case records text by graph fusion;secondly,based on trie tree and parallel computing technology,an acceleration method is implemented,and a high-performance record-oriented dependency parsing model is built;thirdly,based on entity recognition,entity relation extraction by dependency syntax,the transformation from unstructured recorded text to structured case elements is proposed through global relation reasoning;lastly,based on the knowledge graph,combined with the case element information,the record is expressed intelligently,so that it not only conforms to human cognition,but also supports computer analysis and processing.There are three major contributions of this paper as follows:(1)Proposing DAT-MT accelerated graph fusion dependency parsing model,which realizes effective fusion and high-performance parsing of dependency parsing model constructed by cross domain and imbalance corpus: Through the weight optimization of the edge of the graph,the graph fusion method integrates the spoken and specialized syntactic features contained in the small size tree bank of telecom fraud cases into the written language dependency parsing model constructed by public large-scale annotated news text corpus,which improves the accuracy of the model in generating the syntactic tree on case record text.The DAT-MT acceleration method introduces Double-Array Trie and Multi-Threads for the graph fusion dependency parsing model,which makes speed of model analysis increases several times,and has ability of realtime analysis and processing massive case records;(2)Proposing a method of extracting case elements based on dependency syntactic analysis,which realizes the transformation from unstructured telecom fraud case records to structured case elements: The recognition and location of various entities in unstructured text are realized by question answering unit classification,named entity recognition and regular expression technology.Based on the semantic information obtained from dependency parsing,this paper extract many kinds of entity relationships,and transform entity to case elements by global entity relationship reasoning.(3)Proposing a knowledge expression method for records of telecom fraud cases,which can express the records as knowledge which not only conforms to human cognition but also supports computer intelligent analysis: Based on RDF model,the structured case elements are abstracted into knowledge,and OWL is used to build the knowledge graph model of case records;the spatial semantic of the knowledge graph of record is expanded to support the spatial analysis and knowledge reasoning by address analysis and standardization from POI data;the hot spot identification and the construction of space precise anti deception propaganda strategy are realized,and the rationality and practical value of knowledge expression method are verified by spatial clustering and knowledge query.