Natural language parsing of patient complaints in Indonesian language

Abstract

In Indonesia, patient complaints are recorded in the form of free-text data or a narrative text by the doctor when taking the medical history or conducting the medical interview. This text, although recorded in electronic medical records (EMR), is difficult to process computationally because the computer does not recognize natural language. The structure of the Indonesian language differs from that of English. Moreover, the language of patient complaints is structured differently from the Indonesian language in general. It does not consist of the S-P-O-K (Subject-Predicate-Object-Adverb) structures that are used in Indonesian sentences. Moreover, there is a wide range of local languages in Indonesia. Based on data on patient complaints obtained from physicians, this study develops production rules for mapping patient complaints. The aim of the study is to develop a parsing method that automatically maps patient complaints from an unstructured text into a structured text that can be recognized by the computer. In the parsing process developed in this research, a narrative text that has been split into words and/or separated phrases/clauses is used to conduct a suitability search of the lexicon. The lexicon that exceeded the minimum suitability value (threshold) and the highest (maximum) suitability value was selected as the candidate for the lexicon. This study was conducted with consideration for the important information in the free text of patient complaints and could be used subsequently to support a wide variety of clinical decisions.