Sentiment analytics vs emotion analytics: Comparing and contrasting their customer insight valueby
Emotion and sentiment analysis are two distinct methods that yield two different types of insights. So how do they differ and how can they help us understand customers?
There are plenty of arguments to say that failing to anticipate, recognise and respond early to customers’ emotional states can leave them feel scared, frustrated, powerless and ignored. This article tries to uncover what emotions are, how they are different from sentiments, and why understanding and reacting to emotions is important in customer experience management.
The concepts of ‘emotion’ and ‘sentiment’ have traditionally been interchangeably used; however, we must distinguish one from another. Producing a working definition of the concepts is a pre-requisite to discovering their underlying mechanisms to study them better. In turn, this can result in actionable insights for the betterment of both customer and organisation.
So, what are emotions and sentiments?
Emotions versus sentiments
In simple terms, an emotion is an embodied response that occurs when we experience an event, with chemicals being released in our brains that make us feel positively or negatively about the event1. A sentiment, on the other hand, represents an emotional disposition that emerges from our emotions and is developed over time; it is also enduring and generally quantified in terms of polarity (or valence) of emotion (i.e. negative, neutral, or positive).
For example, if a customer says “I am sad”, then it is understood that the emotion felt is ‘sadness’, while the sentiment behind it is ‘negative’. Similarly, ‘hope’ and ‘joy’ are two different emotions, while they are both positive sentiments. Of course, having two positive emotions does not necessarily mean that they have the same intensity (also known as arousal).
Emotions can be viewed along various dimensions. Theoretically, there is no limitation to the number of dimensions we can consider. One of the initial views places emotions in a two-dimensional space, wherein one dimension is represented by valence (or how negative or positive a stimulus is) and the other dimension by arousal (or how mild or exciting a stimulus is, ranging from low to high). For example, happiness is considered to be moderately arousing with positive valence, while anger is highly arousing with negative valence.
However, most of the studies nowadays use three dimensions. Such is the case of the valence-arousal-dominance (VAD) model (Figure 1). Let us view this model on a continuous scale, wherein valence ranges from positive to negative, arousal ranges from low to high, and dominance ranges from submission (low dominance) to feeling in control (high dominance). In view of this model, fear, for example, has negative valence, high arousal and low dominance.
Figure 1. A three-dimensional space of emotions: The VAD model.
One aspect we should emphasise at this stage is that the above definitions imply that, traditionally, emotions have been characterised as having an episodic nature. The fundamental idea behind emotions has been that a noticeable change in the functioning of the person’s body (e.g. blushing, increased heart rate, and so on) is brought about by a triggering event that can be external (e.g. the behaviour of others) or internal (such as one’s own thoughts or sensations). Also, that the emotion episode lasts for a very brief moment (generally, less than one second), and that it is normally easier to identify the onset than the offset of the changed state. There are also characteristics that all emotions share, i.e., rapid onset, short duration, unbidden occurrence, automatic appraisal, and coherence among responses.
However, if one considers the presence of ‘emotional states’, then one implicitly adheres to the existence of a relative stability of emotions over time. And this is highly important. Because this implies that, for example, the impressions left by service providers are long-lasting and can heighten the impact of a service experience, for better or worse. Emotions are also ‘relevance detectors’, which means that once customers feel an emotion about an entity, they know that it has relevance for them.
So, there are plenty of arguments to say that failing to anticipate, recognise and respond early to customers' emotional states can leave them feel scared, frustrated, powerless and ignored.
Emotion experience is not what happens to a customer; it is what a customer does with what happens to him.
The existence of “emotional states” suggests, therefore, that emotion experience is not necessarily self-contained within the moment within which an event happens to someone. Instead, it extends beyond that moment and comprises the lessons that the person learns and applies from that moment.
To put this in context, if a customer has a bad sales experience with a company, that customer will likely be more wary of interactions with the company on future occasions, or at least, be more cautious when it comes to purchasing other products or services.
Hence, if the company cannot adequately anticipate those negative customer emotions on time, it will not be able to adjust its offerings to mitigate them. A disastrous consequence, especially for high-emotion services – those that trigger strong feelings before the service even begins, for example, services such as home buying, airline travel, or services relating to major life events such as birth, marriage and death.
The general rule is quite simple: if the company can unearth customer emotions, it can do more of what they like and less of what they hate.
Previous CX research concluded that emotions play a significant role in customer decision-making. For example, studies have shown that positive CX emotions are linked to positive outcomes (such as repurchase behaviour), while negative emotions lead to negative outcomes (for example, customer dissatisfaction).
However, research has also shown that excessive positive emotions can translate into overconfident decision-making, leading to negative outcomes. Just like a negative emotion does not necessarily mean that the entire customer experience is negative. There are also studies that show that positive and negative emotions can co-exist; for example, a customer may feel happy about a purchase, but unhappy about the price paid for the same.
With mixed research results reported in the extant literature, it only makes sense to analyse each company and its customers in their unique context. But the idea is that independent of the level of association or correlation between customer emotions and outcome, emotions have an impact on customer decision-making and moreover, such impact is quantifiable. And this is a good start.
Emotion versus sentiment analytics
There is a wide range of models and approaches available to analyse and quantify emotions and sentiments. The nicety of it all is that one does not need to ask customers directly about how they feel; instead, emotional experiences can be ethically mined from customers’ own descriptions of instances of critical significance from their customer journey - also known as critical incidents or simply put, ‘moments that matter’.
Sentiment analysis, or opinion mining, is an area of natural language processing (NLP) that aims to extract people’s opinions and views towards specific topics. As previously indicated, sentiments are quantified in terms of polarity (that is, how positive or negative they are). Hence, a sentiment model generally includes three classes (i.e., positive, negative, and neutral), and a polarity value within a given range (e.g. from −1 to +1, with scores of −1, 0, and 1 for negative, neutral, and positive, respectively).
Over the last decade, significant efforts have been made to explore various aspects of polarity and sentiment content of data, such as classifying customer reviews as positive, negative, or neutral, detecting subjective or objective reviews, and visualising sentiments in written reviews. But that is pretty much all that there is. When it comes to exploring emotions, efforts have been traditionally limited.
Emotion models are more complicated, which may explain the scarcity of studies. The complexity is because emotions are multidimensional - theoretically, with no limitation to the number of dimensions. Additionally, there are various emotion classification frameworks. For example, Eckman’s model identifies six categories of basic emotions: anger, disgust, fear, surprise, sadness, and joy (Figure 2), while Plutchik’s ‘wheel of emotions’ identifies eight primary bipolar emotions: anger, anticipation, joy, trust, fear, surprise, sadness and disgust (Figure 3).
Figure 2. Ekman’s six basic emotions. Clockwise, from top left: anger, disgust, fear, surprise, sadness, and joy.
Figure 3. Plutchik’s wheel of emotions.
Therefore, while the number of classes is still three (i.e. positive, negative and neutral), each class is further divided into emotion categories, the number of which depends on the framework used. Then, these categories receive an emotional score.
If we consider that the overall range is from −1 to +1, then all categories of positive emotions (such as happiness) will range between (0, +1), while all categories of negative emotions (such as sadness) will range between (−1, 0). Text mining techniques can be used to derive emotion scores of different valence and different levels of arousal, as well as different levels of dominance, which in turn can be fed into behavioural predictive models to improve their performance.
Consequently, emotion and sentiment analysis are two distinct methods that yield two different types of insights. Emotion analysis adds an extra level of granularity when compared to sentiment analysis. Such granularity works to increase computational complexity. On the positive side, however, the depth of insights that can be obtained is greater.
It can be easily noted that if we know the emotions and their scores, we can also know the sentiments and their scores. But the inverse is not true. In this sense then, emotion and sentiment analysis models can be viewed as complementing each other for CX success, with emotion analysis as an additional layer on top of the relatively more simple-to-perform sentiment analysis.
Ethically mining customer emotions can enhance customer experience, which in turn can drive a new wave of profitability for the company. Formulating and embracing an emotional connection strategy, however, entails deep customer insights and expert analytical capabilities. But, if done properly, it can be a source of real competitive advantage and growth. So, the payoff is huge.
To find out whether you can use emotion detection to improve your customer or employee experience and drive improved business results, contact us at www.anthrolytics.io/contact-us
1 Research has suggested using a design feature approach to distinguish between emotions and other affective phenomena, such as mood, interpersonal stances, sentiments, attitudes, and personality traits. This design feature approach means differentiating the different states by their relative intensity and duration, the rapidity of state changes, and the extent to which the state affects open behaviour, among others.
Considering the design feature approach, an emotion has been formally described as a “relatively brief episode of synchronised response by all or most organismic subsystems to the evaluation of an external or internal event as being of major significance (e.g., anger, sadness, joy, fear, shame, pride, elation, desperation)”.
Charles has a PhD in Operations Research. He is a leading AI and Data Science voice and a Behavioural Predictive Analytics enthusiast. He has published more than 150 research outputs with strong scientific rigour. Skilled in the art of monetising data, with a strong sense of data ethics, and a demonstrated ability to leverage data assets for...