Research On Key Technologies Of Knowledge Graph Costruction For The Knowledge Field Of Ship

Compared with traditional information management methods,the knowledge graph with its powerful semantic processing and open interconnection capabilities can help people quickly sort out the logical relationship between target knowledge and have a good effect on the realization of intelligent reasoning based on knowledge.Compared with the general knowledge graph,the domain-specific knowledge map is usually used for various complex auxiliary analysis or decision support due to its depth and completeness of knowledge,richness and strictness of data models,and high accuracy of description.On the basis of investigating and analyzing the existing key technologies for knowledge map construction,this thesis relies on the construction of knowledge maps in the ship's knowledge domain to explore key technologies such as named entity recognition,relationship extraction,and knowledge fusion in the construction of knowledge graphs in specific fields,the main contents of this article include:(1)Aiming at the problems of nesting and long length of named entities in the ship's knowledge domain,a named entity recognition algorithm based on word vector cascade model is proposed.Firstly,sequence labeling and sequence correction are completed through high-level and low-level network structure.Finally,the upper level output sequence labels are calibrated and the named entity recognition results are output by conditional random fields.Experimental results show that the proposed model based on word vector has achieved good results in complex named entity recognition,and its F1 value(F1 value is weighted harmonic average value)reaches 87.93%.(2)Aiming at problems such as the lack of factual corpus in the extraction of knowledge areas for ships and the inability of deep neural networks to learn high-level data features,according to the characteristics of text data in this area,this thesis proposes a method for extracting mixed relationships based on rules and trigger words.For semi-structured text data A rule-based approach is used,using regular modules and dependent syntax techniques to complete data relationship extraction.For unstructured text,a trigger word-based extraction algorithm is used to match the text vocabulary with the words in the trigger word dictionary to obtain the corresponding relationship type.Finally,more than five kinds of entity relationships are extracted,and the extraction effect is good.(3)Aiming at the problem of error accumulation in relation extraction in the field of ship knowledge,this thesis fuses entity recognition and relationship extraction models.Through this entity-relationship joint approach,the end-to-end learning of original corpus-to-result extraction can be achieved,which can effectively reduce the problem of error accumulation and provide a good user experience.(4)Aiming at the problems of severe redundancy and heterogeneity of the extracted knowledge,this thesis uses an entity alignment algorithm that represents the learning knowledge graph to perform knowledge fusion,that is,the vector similarity calculation is used to complete the knowledge fusion,ans the neo4j graph database is used to complete the storage.In this thesis,through in-depth research and improvement of key technologies for the knowledge graph construction in the knowledge field of ships,the accuracy of named entity recognition and relationship extraction and F1 value are improved,and knowledge fusion such as entity alignment is performed on the redundant knowledge of the graph to build high quality the domain knowledge graph laid the foundation.