Content area
Full text
Introduction
The considerable advancements of social media during the past decade, along with the profusion of digital channels, such as social networking sites (e.g. Facebook), microblogs (e.g. Twitter) and media sharing (e.g. Instagram or Youtube), have revolutionised not only the way brands communicate with their consumers but also the roles of consumers in the marketing process. In a sense, social media gives consumers the same, if not more voice than brands, disrupting marketing processes and creating serious dilemmas and challenges for marketers (Constantinides et al. , 2008). Brand managers can no longer afford to ignore their consumers' important online voice (Gensler et al. , 2013). They are also offered new opportunities to tap into the unfettered consumer-generated content (CGC) readily available on social media platforms. With digital marketing now treated as a "many-to-many conversation" between businesses and consumers as well as among consumers themselves (Lusch et al. , 2010), the traditional one-way business-to-consumers transmissions is becoming obsolete.
A recent trend in the digital marketing analytics sphere is to track and analyse consumers' feelings and opinions about specific brands, products or services attributed to the CGC on social media (Hemann and Burbary, 2013). The objective is to classify positive and negative CGC, typically text-based, according to some manual or automated classification methods. For example, marketers can retrieve timely consumer feedback on a new product by evaluating consumer sentiment expressed in the comments on a Facebook post or in tweets with a specific hashtag related to the product.
Given the large volume of CGC, commonly referred to as "Big Data" that has grown along with the uptake of social media platforms, the qualitative manual analysis of consumers' sentiment conveyed in online brand-related content is no longer practical. To put this into perspective, Twitter generates over 500 million tweets each day, and there are 4.75 billion pieces of content per day on Facebook. This raises the need to develop automated tools for identifying and analysing consumer sentiment expressed in text (Wang et al. , 2012).
Two prominent approaches to automated sentiment analysis exist. Classification using a lexicon of weighted words (Taboada et al. , 2011) is a widely used approach to sentiment analysis in the marketing research community (Bolat and O'Sullivan, 2017), as it does not require...