[New Publication] Approaches to emotion and sentiment analysis

Emotions are central to how we respond to stimuli around us, and consumption is no exception. For instance, the emotions that we are exposed to on Facebook, influence our own emotions and our subsequent posting activity on this social network. With so much of our daily interactions (shopping, leisure, education, conversations, …) taking place online, it is no surprise that marketing academics and practitioners, alike, are interested in the potential of using online platforms to identify consumer emotions, and learn about them (e.g., drivers and consequences of said emotions).

I address the potential and limitations of studying customer emotions online, and cover various techniques to do so, in the chapter entitled “Approaches to emotion and sentiment analysis”, which is part of The SAGE Handbook of Digital and Social Media Marketing edited by Annmarie Hanlon and Tracy L. Tuten. 

In this chapter I start by clarifying the difference between emotions and sentiments; what is meant by emotion vs sentiment analysis; and the importance of looking beyond whether a sentiment is positive or negative, and also consider whether it is active or passive (e.g., gratitude vs happiness; or anger vs frustration). If you don’t know what I mean by sentiment analysis, or valence vs strength of sentiment, you may want to check my introductory post, here.

Then, I review three techniques for identifying and collecting sentiment data, at scale, in the digital environment, and identify their respective advantages and disadvantages. These approaches are:

1. Asking customers about emotionally charged episodes that they experienced in the past, via online interviews or questionnaires; and
2. Placing customers in emotionally charged situations via online experiments; or 
3. Observing customers when they experience an emotionally charged incident via online conversations. 

Subsequently, I discuss techniques for analysing digital sentiment data. I start by emphasising the importance of pre-processing the data. Then, I cover the most popular methods in top-down and bottom-up approaches. Top-down approaches are those take the expressions of sentiment as the starting point, and then look for how they vary with the variables of interest (e.g., product features). Conversely, in bottom-up approaches the sentiment expressions are derived from the dataset.

Lastly, I reflect on the challenges associated with emotion and sentiment analysis, both those related to the study of sentiment per se, and those related to the use of technology to automate the analysis process. Some of those challenges are also discussed here.