Türkçe medikal metin ayrıştırma ve sınıflandırma

This study includes approaches to identify the hospital departments with the methods of machine learning by referencing narratives of patients typed by expert physicians in the hospital information management systems. The main challenge is the preservation of semantic integrity while resolving the complex Turkish grammar rules on medical context. A Turkish medical parser was built and was characterized by specific abilities to extract medical entities and to act according to Turkish grammar. This parser identifies the syntactic, morphological, lexical and semantic features on the medical text. This parser basically discovers the misspellings, disassembles the possible morphemes of a term, assigns the optimal part of speech tags, gets rid of the morphemes that do not affect the semantic intensely and then decides to the best parse regarding the language and word2vec models. A medical corpus which includes patients' medical narrative records and hospital departments that corresponds to those narratives has become the basis for these studies. Additionally, a Turkish medical lexicon has been generated to extract medical entities. It consists of clinically favored terms that specify and differentiate diagnoses. These overall parsing steps provide to perform an exhaustive analysis of Turkish texts through the medical and agglutinative linguistic perspective. Secondly, the features extracted by this medical parser were used to predict hospital departments from patients' narrative records. The accuracies of the Multinomial Naïve Bayes, Support Vector Machines and Convolutional Neural Networks are evaluated to classify the medical content. The Support Vector Machines approach has acquired 98.16 % accuracy rate to classify hospital departments from medically parsed patients' narrative text.