Method and Models of Extraction of Knowledge from Medical Documents

The paper analyzes the problem of extracting knowledge from clinical recommendations presented in form semi-structured corpora text documents natural language, taking into account their periodic updating. considered methods intellectual analysis accumulated arrays medical data make it possible to automate a number tasks aimed at improving quality care due significant decision support treatment process. A brief review well-known publications has been made, highlighting approaches automating construction ontologies and graphs problems semantic modeling problem-oriented corpus. structural functional organization system extraction automatic an ontology graph corpus for specific subject area is presented. main stages dynamic updating are considered: named entity extraction, annotation, term keyword topic modeling, identification, relationship extraction. formalized representation texts was obtained using pre-trained BERT transformer model. selection triplets "object" - "action" "subject" based on part-of-speech markup used construct fragments graph. An experiment carried out given (162 depersonalized case histories patients pediatric center) without preliminary order test proposed solution constructing them. experimental results confirms need deeper take specifics documents. For unmarked texts, demonstrates satisfactory performance view atomic included automatically generated ontology.