Research On Key Technologies Of Constructing Domain Knowledge Map

Knowledge is the cornerstone of human civilization,With the progress of science and technology,the knowledge map provides a new means for people to better manage and use knowledge.In recent years,with the rapid development of deep learning technology,it has provided a strong support for natural language processing tasks,making great achievements in the research and application of knowledge maps.Through knowledge extraction,complex information can be transformed into an orderly,operable and visual knowledge triad,and then the domain knowledge map can be constructed by means of knowledge fusion.Based on current technologies and algorithms,this thesis comprehensively analyzes the massive financial text data,and constructs a knowledge map that can provide comprehensive support for the financial industry.This thesis’ s primary focus is as follows:(1)A multi-task learning and pointer network-based entity recognition method is proposed in this thesis.By using multi-task learning method to effectively reduce the cumulative error between tasks,and introducing the pointer network model,combined with How Net knowledge base to design a multi-layer annotation method for nested entities,we finally realize a domain-specific named entity recognition model.Experimentation on both public and domain datasets has been conducted to a satisfactory degree,the results of which demonstrate that the model presented in this thesis is advanced.(2)This thesis presents a multi-headed selection relationship extraction method based on the relative position representation of entities.Introducing a multi-head selection framework,the model combines relative position representation of entities to judge and classify the domain entity relationship,thus uncovering potential information,minimizing the impact of incorrect labels,and Enhancing the precision of the model for extracting domain text.Similarly,Testing of both public and domain datasets has been done to prove the model’s accuracy.(3)In order to build a domain knowledge graph,this thesis first collects data from specific domains,then extracts domain entities,supplements domain dictionaries,builds relation sets,improves relation categories,and then uses models for relation extraction to obtain a complete knowledge set.Finally,through knowledge fusion,we can build,store and visualize the knowledge graph in the financial field.After rigorous system testing,we find that the financial knowledge graph proposed in this thesis performs well in domain specific applications.