This post is an extract from the chapter “Approaches to emotion and sentiment analysis” that I contributed to The SAGE Handbook of Digital and Social Media Marketing, which was edited by Annmarie Hanlon and Tracy L. Tuten.
The study of expressions of emotion and sentiment is a valuable, and increasingly popular activity, not just as a managerial practice, but also as a scholarly one. For instance, see Kumar et al (2021) for a review of popular text mining applications in the services management literature. However, it is not without challenges. In this section, we consider two types of challenges: those associated with the study of emotions and sentiments in themselves, and those associated with the use of technology to analyse emotion or sentiment data.
1. Expressions of emotion and sentiment
Emotion and sentiment analysis concern itself with capturing and measuring expressions of feelings. However, how people react to an emotional trigger, and express their emotion, can vary widely. People across the world reacted very differently to the Covid-19 crisis, and the public health measures adopted by governments, such as lockdowns or the use of contact tracing. The expression of emotions and sentiment can also vary over time, both in terms of the language’s syntactic features and in terms of style. For instance, LOL started by being an acronym for ‘lots of love’, but now is also used as a replacement for ‘laughing out loud’.
These expressions can also vary with where the emotion or sentiment is expressed. For instance, social media users tend to apply certain colloquialisms and abbreviations, that they might not use in an online product review, or an interview. It has also been observed that online reviews are not only mostly positive, but they tend to follow a J-shaped distribution, with many ratings, a few bottom ones, and very few in between.
Moreover, emotions and sentiment may be expressed through subtle elements, such as the use of exception or conditional clauses, or even the choice of words and their placement. Emotions and sentiment can also be expressed through the use of irony and sarcasm. This is particularly prevalent in social media content. An example of this is provided in Figure 3, via a Twitter conversation which alludes to the crisis faced by Samsung, when its Galaxy Note7 phones were found to overheat and catch fire.
Figure 3. Sarcasm and irony in social media conversations
It is also possible for a single segment of text to express more than one emotion or sentiment. For instance, the author of a product review may judge the product positively, but express dissatisfaction with specific features. This type of situation creates uncertainty regarding the dominant feeling. In this example, whether the review should be classified as positive or negative depends on whether the focus of the analysis is the overall impression or the specific features, respectively. While this may be possible to do in large texts, such as interview transcripts, it is very difficult to achieve when analysing very short text segments, as is the case of short answer in surveys, or entries in popular social media platforms such as Twitter or Snapchat, where the short length or duration of the segment may hinder the identification of multiple foci within one segment of text. For instance, the short sentence “The early shift sucks. Oh well at least my latte is yummy :)” captures two different emotions, and refers to two separate objects, even though it only has 13 words, and 48 characters.
Lastly, sentiment about an object may be expressed in an indirect form, such as through comparisons. For instance, the expression “100 copies of Ghosts sold overnight means a definite Starbucks run this morning. Possibly coffee out twice this week! Maybe even sushi!!” lacks any emotionally charged words (e.g., celebrate, success) that clearly indicate whether the person expressing this sentiment is feeling positively or negatively towards the drink. Instead, the researcher needs to draw on contextual knowledge to understand the complexity of meaning in this conversation.
The challenges associated with the expression of emotion and sentiment are summarised in Table 5.
Table 5: Challenges associated with the expression of emotion and sentiment
| Type | Description | Impacts |
| Form | Way of expressing emptions changes with culture, time and platform | Syntax and style |
| Source | Use of exception, conditional clauses, irony and sarcasm to express emotion | Nuance |
| Focus | Multiple sentiments and/or objects mentioned in the same segment | Uncertainty |
| Context | Emption is expressed by comparison or reference to context | Domain knowledge |
2 Automated analysis of emotion and sentiment
The large volume of digital data available (for instance, on social media) and the ability to collect data easily online (e.g., via MTurk surveys) often result in very large datasets for analysis. Thus, increasingly, both managers and researchers turn to technology that enables the automated tracking and analysis of digital emotion and sentiment data.
Using specialist software for data analysis can help with the manipulation of big datasets, and either the application or the generation of codes related to emotions and sentiments. Using data analysis software can also improve the credibility of the research, even if it does not change the rigour of the analytical work done, or the outcome of the analysis. However, the different approaches to analysing emotion and sentiment data, and the various specific methods, are best suited for some types of data and for specific research goals. Hence, it is crucial to carefully assess the suitability of the selected approach and method for the project at hand. Unfortunately, research shows that this choice (for instance, the choice between unsupervised machine learning vs. semantic network, in the case of latent topic elicitation) is often influenced by pragmatic factors, such as the type of software that the analyst has access to, or the experience of the researcher.
It is also evident that, even though software can accelerate the analysis of the data, researchers still need to be actively involved throughout the process – for instance, deciding what data to retrieve and collate, or labelling data for the training dataset.
The researchers also need to carefully verify the accuracy of the classification, as content analysis software has limitations in terms of discerning nuances in meaning, resulting in the partial retrieval of information, only. This validation is particularly difficult when using off-the-shelf analysis tools, as the working of the underlying algorithms are strongly guarded by the commercial organisations that sell these applications.
In summary, while technology can accelerate the process, and enable the analysis of large datasets, there are a number of vulnerabilities, which may affect the researcher’s ability to correctly detected and classify emotion or sentiment in a dataset. These challenges are summarised in Table 6.
Table 6: Challenges associated with the automated analysis of emotion and sentiment
| Type | Description | Impacts |
| Tool selection | Choice of approach and method | Fit with current project |
| Data preparation | Selection of data and creation of training dataset | Accuracy of classification |
| Tool evaluation | Assess accuracy of results | Confidence in the results |
3. Concluding thoughts
Emotions are key to understand, explain and anticipate consumer behaviour. Digital technology, in particular, offers many opportunities for practitioners and scholars to researching, measuring and describing those emotions or sentiments. However, as noted in this section, the analysis of emotion and sentiment is neither a simple nor a straightforward process. Instead, it is a process embedded in nuance, subjectivity and variability.
It should be emphasised that techniques and technology are constantly evolving, being updated and improved. Therefore, some of the problems highlighted in this chapter may soon be addressed by new methodologies or technical solutions. For instance, dictionaries can be improved, and new techniques can be implemented. However, ultimately, the study of emotions and sentiment needs to be guided by rigour, sensitivity, and criticality – from the point of deciding which method to use to collect data, to the validation of the results of automated analysis, and even the reporting.
Ana Canhoto 