Article Preview
Top1. Introduction
Twitter sentiment analysis (TSA) involves sorting out and grouping of articulated texture opinions automatically on various topics. Such an analysis aids in determining the embedded sentiment and the effect of the expressed opinions intended for the readers (Cambria et al., 2017). An important subtasks of TSA then is the identification of opinionated words (sentiment carrying words). SentiWordNet (SWN) 3.0 and other lexicons avail for this purpose (Baccianella et al., 2010), but a major challenge often encountered in TSA is the presence of negation, resulting in lexical ambiguity.
In health care, negation is an essential linguistic element that can change the semantic orientation of a piece of text. Originating from the biomedical domain (Chapman et al., 2001; Mutalik et al., 2001; Morante et al., 2008), negation plays a vital role in either written or spoken natural languages (NLs). Given the frequent use of negation in biomedicine (e.g., Taylor & Harabagiu, 2018), negation is of significance in health care. As biomedical text represents a vast depository of information feeding the healthcare system, the negation data often seen in clinical reports (Chapman et al., 2001) must be further mined to provide valuable hidden knowledge via sentiment analysis, for example, predicting the risk of dispensing certain drugs. Related to biomedical text mining, the work of Mukherjee et al. (2017), who use NegAIT parser for detecting morphological negation, double negation, and sentential negation in medical text is prominent.
To date, there is a scarcity of research focussing on the medical text such as CONLL2010 (Farkas et al., 2010) and *SEM 2012, centering on determining the negation scope (Morante & Blanco, 2012). One of the popularly used corpus for negation scope resolution has been the BioScope corpora, annotated with cues and scopes (Vincze et al., 2008). According to Vincze et al. (2008), one out of eight sentences in existing biomedical text has negation. A growing number of approaches have been explored over the years to address the negation modelling problem, including machine learning (ML) classifier, rule-based, and more recently, deep learning methods.
Figure 1 displays various approaches applicable in identifying negation cue and detecting scope, specifically in the biomedical domain.
Figure 1. Different approaches used in the literature for the negation cue and scope detection in the biomedical domain
In early research, several rule-base negation detection techniques have been developed and used such as NegEx (Chapman et al., 2001) and NegFinder (Mutalik et al., 2001). These techniques can identify negated findings in the clinical context and discharge summaries, but such approaches are simply no longer adequate for complex structures. Consequently, new techniques and tools have emerged to detect negation in the health care context such as DEEPEN (Mehrabi et al., 2015), MEDLEE, and other ontology-based approaches. The rule-based approach is among the firsts to be used on the Bioscope corpus for negation detection and scope resolution. A notable example is the enduring work of Chapman et al. (2001) on NegEx, a simplistic tool that requires the user to provide a list of phrases and expressions that could be negated. This work has been extensively applied in the biomedical domain.
Table 1 details earlier works in the biomedical domain for the negation modelling including the corpus, results, and negation approaches (for cue detection and scope resolution).