Analyzing Sentiments Expressed on Twitter by UK Energy Company Consumers

Abstract

Automatic sentiment analysis provides an effective way to gauge public opinion on any topic of interest. However, most sentiment analysis tools require a general sentiment lexicon to automatically classify sentiments or opinion in a text. One of the challenges presented by using a general sentiment lexicon is that it is insensitive to the domain since the scores assigned to the words are fixed. As a result, while one general sentiment lexicon might perform well in one domain, the same lexicon might perform poorly in another domain. Most sentiment lexica will need to be adjusted to suit the specific domain to which it is applied. In this paper, we present results of sentiment analysis expressed on Twitter by UK energy consumers. We optimised the accuracy of the sentiment analysis results by combining functions from two sentiment lexica. We used the first lexicon to extract the sentiment-bearing terms and negative sentiments since it performed well in detecting these. We then used a second lexicon to classify the rest of the data. Experimental results show that this method improved the accuracy of the results compared to the common practice of using only one lexicon.