Biomedical Named Entity Recognition Using Different Language Models

Abstract

Information extraction in the medical field involves the management of many important functions such as identification of medical terms, identification of qualities such as denial, uncertainty, severity. The whole process is based on some basic NLP process such as making tokens, part of speech tagging, and parsing. The presence of a large amount of domain specific terminologies in biomedical literature makes information extraction in the field a challenging task. An entity is a word or a sequence of words in text which represent the same word. NERinvolves identification of named entities and their categorisation into different predefined categories.The term “Named Entity” (NE) was first used at the sixth Message Understanding Conference (MUC-6)[1]. In the biomedical domain, the named entities can be classified as proteins,genes, drugs, diseases, organs, tissues, etc. Non standard use of abbreviations, synonyms, synchro nizations, ambiguities and the frequent use of phrases in order to describe the entities make NER in the biomedical domain a challenging task [2]. Named En tity Recognition(NER) is a very important task of Natural Language Processing and is considered a building block for various other tasks. In the biomedical do main NER extracts meaningful entities from clinical text and records which are then used for downstream tasks such as relation extraction, entity resolution, etc. Traditionally, NER models mostly relied on rule-based systems or dictionary approaches. Such models are difficult to implement and require contextual knowledge about the biomedical domain.These models have proved to work with a high precision but have shown lower recall. However, these kinds of systems are inefficient in dealing with new or unknown words.