| Knowledge graph is a significant form of knowledge representation,which is considered to be an effective technique to tackle complex issues of multi-entities and multi-relationships by utilizing graph to describe entities and their interrelations in real world.The enterprise knowledge graph(EKG)is defined to a knowledge graph throughout the core business and processes,which is established by enterprises on the basis of their own requirements.By constructing EKG,the processing of enterprise data assets can be better upgraded to knowledge assets in order to serve for the development of enterprise business.Aiming to the actual requirements of asset risk assessments in the process of acquisition and management regarding non-performing assets of an asset management enterprise,this thesis firstly research and construct the company’s own knowledge graph of non-performing assets,so as to offer a guideline for the risk assessments required by the acquisition,management,and disposal of non-performing assets.Furthermore,the essential issues to be solved in the construction of knowledge graph are named entity recognition,relationship extraction,attribute extraction,and so on.Thus,this thesis study these subjects in depth,i.e.,the named entity recognition and the relationship extraction in the construction of the non-performing assets knowledge graph(NPAKG),and risk assessments in the actual applications.The main work and contributions of this thesis are as follows.1.Focusing on the recognition of flat and nested named entities in the construction of the NPAKG,a named entity recognition algorithm based on BERT-CRF & BERT-MLP is proposed.For the flat named entities and external entities in nested named entities,a recognition method based on BERT-CRF is proposed and enhance the recognition accuracy by building the dictionary of enterprise non-performing assets.Moreover,for the internal entities in the nested named entities,the BERT-MLP model is adopted to classify the internal entities.Experimental results manifest that the proposed method can efficiently identify flat and nested named entities in the text.2.In order to solve the issues of the extraction of restricted domain relationship and the discovery of hidden relationship in the construction of NPAKG,a relationship discovery method based on entity embedding & restricted domain relationship network(EE-RDRN)is exploited.In this approach,Bootstrapping is utilized to extract the entity restricted domain relationship,construct the entity restricted domain relationship network and the entity embedding relationship network,and merge to form the EE-RDRN.In light of the EE-RDRN,the possible relationships among attributes are further explored to excavate the hidden relationships among entities.Experimental results show that the proposed method can effectively extract the predefined relationships and discover the hidden relationships among entities.3.On the basis of the NPAKG,an entity risk assessment model based on the entity risk directed hypergraph is further developed to achieve the entity risk assessments.Experimental results illustrate that this presented approach can efficiently calculate the entity risk and provide a guideline for the enterprise investment decision.4.The design and implementation of the prototype system of the NPAKG are described in detail,and the related technologies and practical operation effects of the system are further summarized and exhibited.The main functions of the designed system include: named entity recognition,relation extraction,entity risk assessment,and knowledge graph visualization.The trial operation of the system reaches the expectation. |
