When sentiment analysis began to be used in marketing activities in the early 2000s, the marketing and advertising industry opened up a wide range of horizons. Knowing consumer sentiment allowed for better validation of actions and more precise targeting. Today, 20 years later, another door has opened – instead of just studying sentiment, you can now study emotions, which brings even more benefits. What are the benefits?
FROM THIS ARTICLE, YOU WILL LEARN:
Sentiment versus emotion – how are they different?
But first, learn the definitions of sentiment and emotion – the most important differences come at this level, because sentiment and emotion are two different phenomena.
- Sentiment is a state in which the person experiencing the emotion is able to name what he or she is experiencing, connect that experience with thoughts about his or her experience, and then make a conscious decision toward the source of the stimulus. Sentiment is thus the sum of physiological reactions (from the body) and cognitive processes caused by the experience of an emotion. As it is a conscious state, it can be sustained by the person experiencing it. In general, it is a broadly defined mental attitude toward a given experience.
- Emotions are the physiological and, consequently, psychological reaction of the brain to an external stimulus or an experience connected with such a stimulus. The experience combined with the body’s reaction (hormone action) causes a brief and unconscious state that leads to a specific action, such as starting a fight, fleeing, freezing, delight, disgust, and so on. More importantly, each emotion is associated with a different type of response that results in a different behavior.
So, as you can see, without experiencing an emotion, you cannot experience a sentient state. Emotions operate at the lowest, behavioral level, while sentiment is the processing, reflection, and attitude of the experiencer. Emotion causes spontaneous reactions, while sentiment causes conscious and controlled reactions. And the most important difference – in the primary understanding of emotions alone there are 8 primary emotions (according to Plutchik’s theory these are joy, sadness, anger, fear, disgust, surprise, anticipation, and trust), and after the combination there are secondary, higher, leading to the formation of feelings. In the case of sentiment, we will only speak of positive and negative sentiment, sometimes distinguishing neutral sentiment as well.
Sentiment analysis and emotion analysis – comparison of tool capabilities
Just as the two phenomena differ, so will the tools to study them. These tools use advanced technologies, based on neuro-linguistic programming, machine learning and other AI algorithms. Currently on the Polish market of companies dealing with sentiment analysis in web content you will find several significant and even reputable brands, but in the field of emotion research – in practice – only one, but a pioneer – Sentimenti. Below you will find comparison of general capabilities of sentiment and emotion analysis tools.
ANALYSIS OF EMOTIONS
Conclusion: the popular (even in the industry) use of the terms sentiment analysis and emotion analysis is incorrect. Sentiment is a much narrower concept, indicating in practice only the overtone of an utterance and – possibly – the mood of its author. Meanwhile, emotion analysis describes the level of particular emotions (we get the percentage result of 8 components, plus the type of sentiment and the value of emotional arousal); having such data we can with high probability predict consumer reactions and behaviour.
As you can see, it is a large amount of data that is detailed; its analysis is more difficult to perform, but more accurate and cross-sectional. As Sentimenti points out, the catalog of their algorithm contains as many as 30 thousand words and phrases collected on a group of 22 thousand people, while the algorithm itself was developed in collaboration with the Wroclaw University of Technology and the Brain Imaging Laboratory of the Polish Academy of Sciences.
How do the results of both algorithms look in practice?
It’s time to practice! Take a look at the analysis below, which was subjected to such an opinion:
I will never buy anything at stinky V(…) again. I recently decided that I do not want to buy tragic clothes in chain stores, so I bought it there and the condition was supposedly perfect, with no sign of use, but with a stain that you can see. The woman wrote me that when she sent it back, it was not there.
This opinion is authentic and taken directly from the portal. What do you read from it when measuring sentiment and what emotions?
Results: anger – 70%, fear – 45%, anticipation – 24%, surprise – 57%, trust – 14%, sadness – 60%, disgust – 62%, joy – 10%, positive sentiment – 10%, negative sentiment – 70%, emotional arousal – 72%.
The statement is a comment from a disappointed customer who, instead of making a purchase in one of the popular boutiques, decided to reach for (in her opinion) better quality second-hand clothes. The purchase turned out to be wrong and the clothes were stained, although they were supposed to be in good condition. The shopper is definitely agitated (72% agitation), angry (high level – 70%), she also feels disgust (62%) and sad (60%) – after all she expected a good purchase. Note that the woman’s expression of emotion is quite high in surprise (57%) and fear (45%) – when you add in the disgust score, you understand that this is her physical reaction to experiencing the scam she fell victim to.
Now look at the same statement in terms of sentiment analysis: you get two results here – negative (70%) and positive (10%). The negative sentiment in the comment is obvious even without automatic analysis of the statement in question – this is evidenced by the phrases stinky, tragic, stain, granny. The phrases condition supposedly perfect and with no sign of use build a low score of positive sentiment.
The most important information for you, however, is hidden – the phrase stinky refers directly to the shopping platform, tragic to the quality of clothes from chain stores, and grandma – to the person selling it. In practice, the tone of the entire statement is set by epithets that are not directly related to the purchased goods, yet you still get a negative result from the sentiment analysis.
Conclusion: Having the percentage scores of basic emotions and knowing the typical physiological reactions corresponding to them, you can estimate the consumer’s behavior in this situation. Anger is equated with an attack response (hence the portal comment), disgust and fear are equated with fleeing, and sadness is equated with a frozen response. An upset consumer is likely to move to another platform or decide to shop in person. She will probably not use this portal again.
Emotion analysis is not only effective in content marketing
You’ve just learned how to interpret the results of sentiment analysis and emotion analysis on an individual example and the range of information you can gain from them. As you’ve also probably noticed, emotion analysis is a much more complete, cross-sectional, and customized solution. But are algorithms for exploring emotions in content only applicable in marketing, piracy or consumer service?
Definitely not. With the development of machine learning technologies and the implementation of better and better artificial intelligence algorithms, the possibilities of emotion analysis are expanding to other industries. Emotions can already be studied, for example, for regular forecasting of stock market prices or investment opportunities of the cryptocurrency market.
Perspectives on the development of emotion analysis tools
If the above information has not yet convinced you of the superiority of emotion analysis over sentiment analysis, then take a look at the development prospects of the former. The possibilities of artificial intelligence are already powerful. Currently, AI has been implemented not so much for research, but for creating emotionally charged content so that you can achieve your goals with it.
These goals include increasing conversions from marketing efforts, acquiring leads more effectively, or creating a customized (sic!) customer experience that takes into account potential crises and how to avoid/bypass them. This last goal definitely improves User Experience and generates greater customer loyalty to the brand. This, in turn, leads to a better image of the brand, and thus its position in the market. So much for technology development perspectives. Your sentiment analysis fails to do that.