Social media sentiment analysis: lexicon versus machine learning

Purpose With the soaring volumes of brand-related social media conversations, digital marketers have extensive opportunities to track and analyse consumers’ feelings and opinions about brands, products or services embedded within consumer-generated content (CGC). These “Big Data” opportunities render manual approaches to sentiment analysis impractical and raise the need to develop automated tools to analyse consumer sentiment expressed in text format. This paper aims to evaluate and compare the performance of two prominent approaches to automated sentiment analysis applied to CGC on social media and explores the benefits of combining them. Design/methodology/approach A sample of 850 consumer comments from 83 Facebook brand pages are used to test and compare lexicon-based and machine learning approaches to sentiment analysis, as well as their combination, using the LIWC2015 lexicon and RTextTools machine learning package. Findings Results show the two approaches are similar in accuracy, both achieving higher accuracy when classifying positive sentiment than negative sentiment. However, they differ substantially in their classification ensembles. The combined approach demonstrates significantly improved performance in classifying positive sentiment. Research limitations/implications Further research is required to improve the accuracy of negative sentiment classification. The combined approach needs to be applied to other kinds of CGCs on social media such as tweets. Practical implications The findings inform decision-making around which sentiment analysis approaches (or a combination thereof) is best to analyse CGC on social media. Originality/value This study combines two sentiment analysis approaches and demonstrates significantly improved performance.