Portrait Analysis Of Key Person For Historical Files

Under the background of the era of big data,the traditional mode of key person control has been challenged.With the development of internet technology,the phenomenon of missing control and off control occurs from time to time in the control work of key person.And information barriers exist between departments,resulting in a large amount of accumulated data didn’t exert its due value.In view of the above problems,this thesis takes the intelligent integral early warning system for key person as the background for the accumulated historical files,combines the theory of user portrait,uses the big data mining techniques to construct and analyze the portrait of key person to assist domain experts in analysis and decision-making to achieve dynamic control of key person.The main research contents are as follows:(1)Key techniques for construction of the portrait of key person.This thesis proposes a multi-dimensional representation framework to guide the construction and analysis of the portrait of key person.An entity recognition model based on cascaded neural network is designed to construct social dimension and solve the problem of entity nesting and writing process in historical files.A key extraction model incorporating thematic information is used to construct thematic dimension,which is to improve thematic relevance and coverage of extracted keywords for historical files.(2)Key techniques for analysis of the portrait of key person.This thesis proposes a representation method of natural dimension,social dimension and thematic demension in the portrait of key person for subsequent analysis.Based on the business analysis requirements of domain experts,the individual and group target analysis based on the portrait of key person is proposed.A random forest algorithm based on under-sampling is proposed for individual target analysis to solve the problem that poor high and medium risk degree determination due to unbalanced dataset of key person.Considering the difference in the contribution of different dimensions to the degree of outlier for the key person and imporving the speed of calculating the outlier degree of each key person,a fast local outlier factor algorithm based on multi-dimensional fusion is proposed for group target analysis.(3)Design and implementation of key person portrait system.The system includes system management platform,data acquisition subsystem,data governacnce subsystem,portrait constrction subsystem,key person warning subsystem and big data computing platform.This thesis focuses on the design and implementation of the subsystem of portrait construction and the subsystem of key person warning,and shows the construction effect of the system.