Abstract
Sentiment analysis or Opinion mining is a widely studied research area in the field of Natural Language Processing (NLP) that involves the identification of polarity (positive, negative or neutral sentiments) of the text, usually done on shorter and emotionally charged text, such as tweets and reviews. Parliamentary debates feature longer paragraphs and a very esoteric speaking style of Members of the Parliament (MPs), making them much more complex.
The aim of the paper was to explore how and if lexicon-based approaches can handle the extraction of polarity from parliamentary debates, using the sentiment lexicon VADER (Valence Aware Dictionary and sEntiment Reasoner) and the Liu Hu sentiment lexicon. We performed sentiment analysis with both lexicons, together with topic modelling of positive and negative speeches to gain additional insight into the data.
Lastly, we measured the performance of both lexicons, where both performed poorly. Results showed that while both VADER and Liu Hu were able to correctly identify the general sentiment of some topics (i.e., matching positive/negative keywords to positive/negative topics), most speeches themselves are very polarizing in nature, shifting perspectives multiple times.
Sentiment lexicons failed to recognise the sentiment in parliamentary speeches that might not be extremely expressive or where a larger sum of intensity-boosting positive words are used to express negativity. We conclude that using lexicon-based approaches (such as VADER and Liu Hu) in their unaltered states alone do not suffice when dealing with data like parliamentary debates, at least not without any modification of lexicons.