Medical Domain Taxonomy Learning Based On Background Knowledge

At present,although the practicality of domain ontology has been widely recognized in the field of biomedical research,there are still many obstacles to its effective use.Domain ontology aims to cover the concepts and relationships in the field as comprehensively as possible.However,constructing the ontology by hand is usually time-consuming,prone to errors,and poses problems of poor updating and poor consistency.Ontology learning as an ontology automatic construction technology can well solve the above problems.Natural language processing,information extraction,information retrieval and machine learning technology advances,promoting the development of ontology learning technology.Ontology learning includes terminology extraction,concept extraction,taxonomy extraction,non-taxonomy extraction and axiom extraction.Taxonomy as an important part of ontology construction,is the focus of research at home and abroad.This thesis proposes a context-oriented text-based domain taxonomy automatic learning scheme,the main process is as follows:(1)Domain terms are extracted from free texts,and the extracted terms are combined with domain background knowledge RxNorm to form concepts.(2)Calculate the five dimensions similarity based on the lexical similarity and the semantic similarity between the formed concepts,including the Levenshtein similarity between concept strings and term strings,semantic similarity based on background knowledge of domain knowledgebase SNOMED CT,and semantic similarity based on background knowledge of Linked Life Data DBpedia.(3)Hierarchical clustering of the extracted domain concepts to complete the integration of multi-dimensional information to obtain the initial concept of dendrogram.Then pruning the tree and labeling the clusters.Finally,we derive the taxonomic relations.Based on the above method,this paper selected the MEDLINE and DRUGBANK corpus to learning the taxonomies,extracted a total of2547 concepts,mapped to 16 top level hierarchies.Then,evaluated and summarized the taxonomic relationships,realized the discovery of taxonomic relationships from the free text.On this basis,ontology modeling of the concept and taxonomies of the extracted domain is completed,and the ontology construction of the corresponding domain is completed.The ontology can be used as a reminder for clinical medication to provide the doctor with the upper-level relationship and underlying relationship of the drugs used,so that the doctor can fully grasp the drug information.