A Knowledge Retrieval And Question Answering System Of Railway Facilities And Geological Factors

With the explosive growth in the scale of railway facilities and geological factors description text data,the performance of information retrieval has a significant impact on the effective use of remote sensing geological factors text resources.In the era of big data,the limitations of traditional search engine retrieval strategies in terms of word expansion and deep semantic understanding make information retrieval face serious challenges.Based on the literature resources of remote sensing geological survey of railways,this thesis discusses the reliability of text semantic vectorized representation,the necessity of text structuring to improve the performance of information search service,and the inevitability of adopting question answering system as the next generation search engine.Building a knowledge retrieval and Q&A system for railway facilities and geological factors based on relevant research and technology.The research work of this thesis is as follows:(1)This thesis analyzes the similarities and differences between full-text retrieval and semantic retrieval from the perspective of principles and technical methods,and verifies the transformation of semantic representation from symbol to vector will help to improve the retrieval performance.Based on the construction of domain thesaurus,a full-text retrieval system supporting word matching and a semantic retrieval system supporting document semantic vector calculation are implemented based on Elasticsearch and Doc2 vec respectively.The result shows that semantic retrieval can effectively achieve word expansion and can more accurately grasp the user’s query intent.(2)Set the ontology of railway facilities and geological factors,and select some domain literatures for manual annotation.Completing entity and relationship extraction of railway remote sensing geological knowledge using deep learning to realize the structured text information.The RDF triples are used to represent the knowledge,and the extracted knowledge is stored in the database after processing operations such as correction and entity linking to build a small domain knowledge base.(3)This thesis realizes the entity and attribute recognition in question sentences based on Bi LSTM-CRF and BERT,completes the understanding and analysis of simple question sentences,and generates structured query language by filling slots with templates.A simple question answering system based on remote sensing geological knowledge base of railways is constructed.