Chapter 7 - An annotation system of a medical corpus using sentiment-based models for summarization applications

Abstract

The amount of information available is growing exponentially every day, especially in the healthcare domain. In order to gain access to this information, an automated annotation system is essential. This system may assist in understanding relevant queries or medical concepts and their related information from the raw medical corpora quickly and accurately. However, a structured corpus is important to build a domain-specific annotation system for healthcare. In this chapter, we are motivated to prepare a structured corpus using our developed sentiment-based medical annotation system. In addition, we have focused on developing a sentiment-based summarization system that assists in understanding the important observations of the corpus. Hence, first we have collected an unstructured corpus from an online resource, namely MedicineNet. Secondly, we have employed WordNet of Medical Events that is presented as a lexicon specific to the medical domain along with machine-learning classifiers to identify and assign medical concepts and their sentiments, respectively. Medical concepts have been identified in the form of uni-gram, bi-gram, tri-gram, and more than tri-grams. Additionally, we have tagged these identified medical concepts by three primary sentiments such as positive, negative, and neutral. Thereafter, we have recognized important parts of the corpus using sentiment-based ranking as its extractive summary. Finally, we have evaluated both the annotation and summarization systems in the presence of a group of medical practitioners.