by Sentimenti Team | Dec 7, 2021 | Conferences, Scientific publications
Place of publication:
Conference: 2021 IEEE International Conference on Data Mining (ICDM)At: Auckland, New Zealand
Title:
Learning Personal Human Biases and Representations for Subjective Tasks in Natural Language Processing
Authors:
Jan Kocoń, Piotr Miłkowski, Damian Grimling, Marcin Gruza, Kamil Kanclerz, Przemysław Kazienko, Julita Bielaniewicz
Abstract:
Many tasks in natural language processing like offensive, toxic, or emotional text classification are subjective by nature. Humans tend to perceive textual content in their own individual way. Existing methods commonly rely on the agreed output values, the same for all consumers. Here, we propose personalized solutions to subjective tasks. Our four new deep learning models take into account not only the content but also the specificity of a given human. The models represent different approaches to learning the representation and processing data about text readers. The experiments were carried out on four datasets: Wikipedia discussion texts labelled with attack, aggression, and toxicity, as well as opinions annotated with ten numerical emotional categories. Emotional data was considered as multivariate regression (multitask), whereas Wikipedia data as independent classifications. All our models based on human biases and their representations significantly improve the prediction quality in subjective tasks evaluated from the individual’s perspective.
Link: ResearchGate
by Sentimenti Team | Sep 30, 2021 | SentiBrand
The study of emotions contained in content is today eagerly used, among other things, to analyze and predict consumer behavior. No wonder. Reading the motivation of internet users helps to better define the target group of a given product or service and then facilitates more precise construction of marketing communication. One of the parameters that should be taken into account when analyzing the emotions contained in the text is emotional arousal. Why? What benefits do we get from it? Check it out in this article!
What should you know about emotional arousal to begin with? Let’s start with a definition. In a previous article, we pointed out that emotion is an experience and also an affective stimulus, causing an action to be taken.
One component of this experience is emotional arousal (arousal).
This arousal reflects the degree of activation of the central nervous system. It directly affects the way (intensity) of feeling a particular emotion. However, it is not a state that is always reflected in the dynamics of a particular person’s behavior. It can lead to increased liveliness, but it can also cause apparent calmness, indifference or even its withdrawal.
Interesting fact. The nature of the emotion experienced has no effect on the value of emotional arousal. Sadness, disgust, or joy can have the same arousal value.
Emotional arousal in terms of content emotion analysis means the level of intensity of a given reader’s emotions towards a particular event or information read. It can be zero, in which case we speak of indifference. It can also be measured and felt as strong: for example, excitement or agitation.
Emotional arousal is a state whose symptoms can be observed in the body. Noticeable physiological reactions include:
- elevated heart rate,
- more frequent and more intense heartbeat,
- accelerated breathing,
- increased perspiration,
- goose bumps on the skin,
- dilation of the pupils and accommodation of the eyes to see at a distance,
- dilation of blood vessels in the external genitalia.
These symptoms may be accompanied by other, less obvious visceral reactions, that is, reactions of internal organs:
- due to the activity of the adrenal glands, adrenaline is secreted into the blood,
- the liver releases glucose into the bloodstream,
- intestinal peristalsis is reduced and bronchial tubes dilated.
As you can see, this is atavism. The so-called reptilian brain is at work here. In the body, mobilization occurs, leading to an immediate reaction to the occurring changes. For example, in the case of fear it will be flight, and in the case of anger or rage – aggression, fighting.
Unlike sentiment analysis, emotion analysis can indicate the level of emotional arousal in statements or other types of content published by internet users. For analysts, communications professionals, marketers and even salespeople, this is a very important factor. Why?
Measuring arousal, sometimes called the temperature of an utterance, gives those measuring it the ability to determine the overall strength of all emotions. It additionally conveys two pieces of information:
- it indicates the severity of the tendency to perform the response attributed to a particular emotion,
- and, at the same time, it shows the magnitude of the obstacles that may block a person from taking a particular action.
The analyst thus gains knowledge of the person’s propensity to act. If, in addition, he knows what emotion is behind the statement, he will also know what action to expect. He will also learn about the extent of the blockages inhibiting the person. With knowledge of the emotions involved, he will be able to more easily determine whether it is an external obstacle or rather a mental block.
What do we gain by having our level of emotional arousal examined in conjunction with a given emotion?
First, we gain the ability to identify the emotional state that results from the content examined, in terms of its intensity
Thus, we distinguish between:
- mood – when the intensity of the emotion is low but can last for a long time,
- emotion – sudden and short-lived emotional states,
- affect – a sudden physiological state of short duration but high intensity, ending in passing and a phase of weariness.
Second, when combined with a given emotion (or compilation of emotions), the analysis will make it easier to determine the source of that emotion
Depending on the type of emotion, the arousal will be sensory or internal, visceral (i.e., aggregated due to the body’s biological imbalance, deprivation of needs, etc.).
And we are only talking about emotions in a primal, animal context! When we begin to operate with derivative, complex emotions, a more complicated and simultaneously more desirable picture will emerge. For we will enter into the so-called reasons of reason, emotions of a transgressive, reflexive character.
Their level and intensity may even give the researcher an idea about the system of values adhered to by the commentator, its violation or realization. It is probably not necessary to convince anyone about the importance of such data for their interpreter.
Third, and finally, the juxtaposition of emotion and arousal allows the analyst to estimate the impact of a given piece of information on an individual’s cognitive development
In what ways? First of all, we know that cognitive processes are connected to emotional processes. The latter affect cognitive processes by giving them a sui generis emotional coloration that stems from the emotions that we are experiencing.
The high intensity of a given emotion will therefore restrict cognitive processes, because these processes will be, as it were, filtered through the filter of the emotions experienced. However, positive stimulation (“reward” that introduces a positive feeling in the subject) improves the process, while the expectation of unpleasant consequences slows it down.
The analysis of primary and secondary emotions in combination with emotional arousal will allow the researcher to attribute specific actions to the results obtained. It will give him a specific picture of the structure of needs behind the analyzed statement. As a result, it will be possible to tailor a marketing message that is much more personalized. If we gather a sufficiently large group with similar results (i.e. similar needs), with such data we can begin designing marketing strategies.
Let’s proceed to test the knowledge described here in practice. As a basis for the analysis we will take comments and opinions from the Internet. They were scanned with SentiTool for the emotions they contain. Sentiment and the degree of emotional arousal were examined. We will see the relationship between emotional arousal and predicted action.
Example 1:
Panel prices drama – just like any other store. I will never again buy anything from the damn XX. This floor is crap, and the company that produces this crap is a huge failure.
Message analysis results:
- anger – 51%,
- fear – 31%,
- anticipation – 30%,
- surprise – 52%,
- trust – 20%,
- sadness – 45%,
- disgust – 42%,
- joy – 19%,
- positive sentiment – 18%,
- negative sentiment – 45%,
- excitement – 59%.
In the results obtained, the high scores of anger, disgust, sadness and surprise are immediately striking. However, these are obvious in a consumer disappointed with a purchase. These values are supported by a strong negative sentiment, which hints that this state will persist and can be nurtured in the way people harbor resentment.
The emotional arousal of the Internet user is high here and will certainly lead to impulsive actions. What are they? The anger factor (51%) may indicate a desire to take retaliation, probably in the form of leaving aggressive, unfavorable comments and opinions online and among people familiar with the author of the post in question.
However, the most important part of the statement is the author’s identification of the “guilty” of the whole situation and, at the same time, the one responsible for his emotional state. And it is … no, contrary to appearances, not the manufacturer of “crap and junk”, that is the company “great failure”. The culprit is a popular DIY store!
We are witnessing an interesting shift of aggression from the manufacturer to the distributor of floor panels. The reason for the transfer is the unattractive price of the purchased product. As a punishment, the customer threatens to give up shopping in this chain. It can be considered that the quality of the product in this statement is of secondary importance. The customer guessed what he was buying and probably had an internal agreement about not the best specification of the product, but he expected a much lower price.
How do we know this? He did some consumer research and compared prices in other stores, he probably checked opinions of other users about given panels. Finally, despite everything, he bought this particular product. The first two sentences of the internaut’s statement (out of three!) refer to the store’s activity, and only the last sentence refers to the low quality of the floor panels. Something tells us that if the author of the opinion had bought the panels in the promotion, the above entry would not have been written at all.
What about the threat of a chain boycott? This boycott will last with the consumer… until they need to make their next purchase. There are a limited number of players in the building and finishing products market, and the aforementioned market is well established and has the fourth largest distribution network in the country. The results of the emotion analysis are not so high in this case as to suggest abandoning the relatively convenient (and, paradoxically, price-competitive) market in favor of seeking similar merchandise in other chains.
Example 2:
Anything better than G… and their latest “toxic masculinity” anti-male campaign, spreading propaganda saying that whites are responsible for all the evil and hate in the world. I will never buy anything from this crappy company again.
Message analysis results:
- anger – 58%,
- fear – 43%,
- anticipation – 24%,
- surprise – 53%,
- trust – 13%,
- sadness – 52%,
- disgust – 50%,
- joy – 10%,
- positive sentiment – 10%,
- negative sentiment – 57%,
- emotional arousal – 64%.
In the message examined, we immediately see a contradiction. The high levels of anger, fear, surprise and sadness are a reflection not of the products themselves, but of the mission or idea that the concern represents. This mission does not agree with the commenting consumer’s vision of the world.
The man is filled with anger (58%). He had largely identified the company with masculinity and suddenly this image began to change. This thesis is supported by high rates of surprise (53%) and fear (43%).
Fear, however, reveals something more to us. It appears because of the blow to the consumer’s fundamental values, which are hidden in the words: whites are responsible for all the evil and hatred in the world. What we get is an image of a conservative person who will not necessarily be appealed to with facts, but who will not mince words when it comes to defending his or her rights or values.
What action can such a consumer take? Anger combined with fear is an explosive mixture – with increased arousal, it can and will lead confidently to an attack. After all, the first step was to post the comment in question. There are bound to be other posts: people with high levels of anxiety are closed to change and are very aggressive in public discussions that undermine their higher values.
Such a person will opt for a consumer boycott (arousal 64%), although we are not sure they will persist in it. Why, you ask? The commentary begins by saying that the campaign is anti-male, showing the toxicity of masculinity. But all the anger focuses on the message of white guilt for the evils of this world. This consumer may nevertheless return to the brand’s products in some time, because he is used to them. He is, after all, masculine.
Example 3:
Irrigation of our greenery is important. At this point I too am going to buy a rotary sprinkler and have already seen several models in the store.
Message Analysis Results:
- anger – 16%,
- fear – 15%,
- anticipation – 41%,
- surprise – 40%,
- trust – 34%,
- sadness – 19%,
- disgust – 13%,
- joy – 35%,
- positive sentiment – 37%,
- negative sentiment – 6%,
- emotional arousal – 48%.
In contrast, the text to be analyzed is seemingly unemotional, neutral. The results of expectation, surprise and trust stand out the most, supported by a not inconsiderable (48%) arousal factor.
What do these results say? The person who left the comment is in the process of buying the product. Heightened anticipation (41%) combined with joy (35%) indicates that she is somewhat excited about the search, checking out the available options and looking out for the optimal offer (again, anticipation).
She is also surprised (40%) but still not frightened (15%) by the number of solutions available on the market. She believes she will find the right device (confidence – 34%). Agitation indicates an advanced process that is likely to end in a purchase.
The statement does not contain many details. You can see from its shape that it is a response to a comment from another person looking for such a product. It starts with a truism and, after a slightly stylistically convoluted construction, quickly moves on to the substance – a declaration of a common goal of the search. Now it is enough to add a link to the relevant product to the entry and… we have a typical whisperer’s comment.
If you would like to know, practitioners of whisper marketing, how your work looks like from the point of view of the study of emotions by artificial intelligence, it looks like this.
Example 4:
I bought such a decanter for my father and am eminently pleased. It is so beautiful that I myself gladly reach for it sometimes just to admire it. Lovely!
Message Analysis Results:
- anger – 12%,
- fear – 14%,
- anticipation – 61%,
- surprise – 50%,
- trust – 65%,
- sadness – 12%,
- disgust – 12%,
- joy – 72%,
- positive sentiment – 75%,
- negative sentiment – 1%,
- emotional arousal – 72%.
The statement is made by an extremely satisfied consumer who chose the product as a gift for his father. We will not find out whether the gift was to the liking of the recipient. Probably neither does the commentator himself. As evidence we have high expectation supported by surprise at 50% and slightly elevated indices of anger, sadness and fear. For the same reasons, we also don’t know if the gift performs well as a decanter.
What we do know is that the purchaser himself certainly enjoyed the purchase. He exalts the beauty of the product, describes holding it in his hands, and we sense the desire to repeat this action. Such emotional state of the commentator towards the product will last – this thesis is supported by our high positive sentiment (75%).
The arousal factor in such conditions can mean a desire to return to the store and buy the same product. Well, unless the son finally gets up the courage to ask his father what he thinks of the gift. There is a chance that he will get this decanter for himself…
The analysis of the comments described above is detailed, almost meticulous. So you may be tempted to respond: This is impractical! What about processing large amounts of data? After all, in order to prepare the communication strategy, monitor the brand image in the network, social listening, advertising campaign, etc. we will need thousands of mentions. How to deal with them?
It’s simple – the analysis of emotions contained in the content is automated. Content is analyzed by AI algorithms operating on the basis of deep neural networks. All mentions or posts are in turn categorized in such a way that a person starting to analyze the data has a simplified task. In addition, the analysis tools used today can be calibrated to the needs of a specific task or use selected classification systems.
So if you need to take your work to the next level with data analysis for marketing, advertising or communication projects, consider analyzing emotions in content.
by Sentimenti Team | Sep 30, 2021 | SentiBrand, Sentimenti research
Emotionality is an extremely important, yet often underestimated part of the human being today. Emotions are a direct heritage of our distant ancestors – they were created to inform us about important events, to protect us from danger, to help us make decisions, to sustain us in action, and to establish and maintain relationships with others. Emotions translate into almost every area of life and without them it is simply impossible to function.
This (perhaps somewhat emotional) introduction is an introduction to a series of articles describing the possibilities of using modern technologies and artificial intelligence to recognize emotions and – in the perspective – to predict future decisions or consumer actions.
The source of information about these emotions can be, for example, posts of the Internet users themselves, published on social networking sites, discussion forums, in the form of comments under articles and wherever there is a possibility to express one’s own opinion. The more initiated will say – this is all about sentiment analysis! And they will be right, but only partially – we will look at these issues more broadly than just through the prism of sentiment. Because emotion is definitely not the same as sentiment, although we often see the two phrases used interchangeably.
What are emotions?
Let’s start by defining emotion. It is an instinctive, spontaneous, but also complex reaction of the brain to a stimulus, both external (perceived by the senses) and internal (memories, imagination or thoughts). Emotions are one of the components necessary for the perception of reality – they participate in cognitive processes, regulate social behavior, and in a healthy person – through mirror neurons – help to understand the emotional state of another person.
Why complex? The result of such a reaction is, among other things, the secretion of hormones and the expression of emotions through the body. Such patterns of physiological reactions as acceleration of breathing, pulse and heartbeat, pain, freezing, tingling or other arousal of the body, muscle tension, nausea, in addition to characteristic facial expressions, body language, sound properties of speech (prosody), etc., may appear in the expression. As a result – after collecting and processing the data, a specific action is taken, e.g. running away, putting up a fight, expressing admiration, appreciation, etc.
In the course of research on emotionality, it has been established (e.g. by Paul Ekman and Robert Plutchik) that some of the emotions occur in people irrespective of their settlement (or upbringing) in a given cultural circle. These emotions are the so-called primary emotions, and among them are: joy, sadness, anger, fear, disgust, surprise (Plutchik developing this theory additionally includes expectation and trust). Emotions can be compared here to colors: in both cases the basic components create their derivatives.
Thus, from the mixture of sadness and surprise comes disappointment, from anger and disgust – contempt, from trust and fear – submission, from surprise and anger – indignation. In a straight line – emotions are a way of expressing more complex constructs – feelings.
What is sentiment?
And how does sentiment relate to emotion? Sentiment can be defined as a peculiar mental attitude that is an aggregation of emotions and related thoughts. It is a decidedly subjective experience and develops from emotion only when we are aware of the presence of a phenomenon. Sentiment combines physiological reactions and cognitive components, and it occurs when we are already able to name the state we are in (or identify, name the emotion affecting us) and, consequently, decide to make a particular response. It will not develop unless it is preceded by the expression of an emotion.
Sentiment is, then, the result and, at the same time, the derivative of many emotions, an emotional state with a longer span of action than emotions – for it lasts as long as we are able to “cherish” its presence within us.
What is sentiment analysis?
Sentiment analysis in the context of Content Marketing is a way to measure and validate the overall impression, the mood that readers are in after reading given content. It doesn’t so much focus on the specific emotions expressed, but rather differentiates the feelings or impressions of the audience after reading the text into “positive” and “negative” (sometimes “neutral” as well) – this is reflected in the “positive” and “negative” sentiments. (sometimes also “neutral”) – this is reflected in the opinions expressed. When analysing the sentiment in the content, we collect data which, although highly simplified, gives us the possibility of quick estimation and easy processing. The analysis of the sentiment contained in the content is a method for measuring the reaction (attitude) of readers after reading, and thus the marketing value of the content itself.
How does this translate into action in practice? Thanks to the sentiment measurement it is easier for the publisher to verify particular articles – to check whether the reception of their message was positive or not; however, a problem may arise in the case of neutral, informative articles and content which will not lead to polarization of opinions when commenting on them. The data collected in this way is simply highly generalized.
Sentiment analysis provides insight into readers’ thematic preferences: it allows you to group content into those that evoke positive and negative associations. It makes it easier for publishers to plan the publication of content and allows for better (albeit not very precise) targeting of content audiences. Remember – the process of forming sentiment is long and complicated, and sentiment itself is susceptible to regulation and control by the reader’s thoughts. As a controlled reaction, it loses its natural dynamics and thus its authenticity.
Emotion analysis and the differences in the two solutions
Using the terms emotion and sentiment interchangeably is a mistake. As we have shown above, an analysis of the sentiment contained in published content is only a slice of a larger picture. A slice that describes a situation in which the reader will control his reactions and, as a result, consciously take action. So you could say that sentiment analysis is based on the logical actions of a given internet user.
Meanwhile, the analysis of emotions is essentially holistic – it focuses primarily on the primal, atavistic element, namely the instinct. Emotions appear spontaneously, they can be very intense and have a short-lasting effect, but they are also a motivator for specific actions. This is why an analysis of the emotions contained in a text, though more difficult to conduct, is in effect deeper and more accurate – it will make it possible to predict readers’ reactions, the actions they will take, or even try to identify the blockages holding them back from specific activity.
Analysing emotions allows you to gather a lot more data on the extent to which readers enjoyed, frustrated or simply found the text boring, which will then translate into more accurate planning of actions towards particular texts and predicting internet users’ reactions: sentiment will show us consumers’ attitudes towards the published content, while emotions – what’s behind these attitudes!
Example: Two reviews of a product are published. One of them says “This product did not meet my expectations,” while the other says “I hate this product with all my being.” Both descriptions will be classified as having a negative sentiment, but they will be significantly different in emotional valence – after all, there is a difference between disappointment or sadness and hatred. The analysis of emotions allows us to detect such, sometimes small, subtleties and indicate the specific types of behavior following them. It will help to guide the employees analyzing the data on what a given consumer may do after issuing such an opinion, or indicate how another consumer, whose product did not satisfy or saddened, will behave.
Let’s look at the results of testing the two sentence examples above with the Sentimenti text emotion analysis tool:
Ten produkt nie spełnił moich oczekiwań
This product did not meet my expectations | Nienawidzę tego wyrobu całym sobą
I hate this product with all my heart
|
| |
| |
| |
surprise / zaskoczenie - 40% | surprise / zaskoczenie - 44% |
| |
| |
| |
| |
For comparison (study conducted with the same tool): sentiment for the first statement included in the sentence – positive – 20%, negative – 25%, emotional arousal – 46%. For the second – positive – 20%, negative – 40%, emotional arousal – 61%.
It is immediately clear that the analysis of emotions gives much more useful information. First comes anger – an obvious emotion, caused by discomfort, dissatisfaction with the purchase, but differentiated into disappointment and hatred. This anger is additionally boosted by fear, which in the second case reaches a significant intensity of 34%. Sadness is also significantly higher in the second statement.
What is not surprising is the similar level of surprise (discomfort in both cases) and anticipation (perhaps of a reaction from the brand or store below the post, or other comments similar in tone). The low level of trust is also not surprising, although it is not close to zero – probably both commenters are waiting for the mentioned reaction from the brand, and they have not experienced a similar situation from its side before. And the joy – you ask? Similar in both statements, it can mean hope for an apology or… satisfaction from expressed anger!
Another issue that makes sentiment analysis less useful is the possibility of data misinterpretation. What are we talking about? The misattribution of sentiment to emotion and the following consequences. An example from a popular fanpage of a well-known large discount retailer: “Why (angle grinder – author’s note) is not available stationary ?”
Conclusion: a consumer comments under a post in Social Media complaining that the product is not available in his local store. From the context of the comment it is clear that the sentiment contained in the comment is negative and the content itself (which can be seen even without the algorithm!) is characterized by sadness. Without analyzing the emotion, it is impossible to conclude that this sadness does not have to have a negative connotation – after all, it shows the brand’s communication tracking, indicates interest in the product and indicates a potential problem with its sale. Valuable information for analysts!
Let’s back this up with another example. We compare two comments and examine them in terms of the emotions and sentiment they contain.
I'm really regretting not being able to buy that blue cheese today that we like so much. Bardzo żałuję, że nie udało mi się dziś kupić tego sera pleśniowego, który tak bardzo lubimy. | Today I saw cheese on the shelf at the store, all moldy, something disgusting.
Dzisiaj w sklepie widziałem na półce ser, cały spleśniały, coś obrzydliwego. |
| |
| |
oczekiwanie / expectation - 31% | oczekiwanie / expectation - 21% |
zaskoczenie / surprise - 44% | zaskoczenie / surprise - 56% |
| |
| |
| |
| |
Sentiment for statement 1: positive – 20%, negative – 35%, emotional arousal – 54%.
Sentiment for statement 2: positive – 4%, negative – 72%, emotional arousal – 64%.
Conclusions. In both cases, sentiment is rated as negative. But if we go into a deeper study (emotion analysis), we see that: confidence level is 2.5 times (250%) higher in statement 1 than in statement 2, joy is 5 times (500%) higher in statement 1, while anger is almost twice (84%) higher in statement 2, sadness is 46% higher in statement 2, and expectation (for something positive) is 47% lower in text 2.
Anger at an intensity of 70% is obvious for the second statement, as are the high scores for surprise, sadness, and disgust. Well, few people would be happy to see such a find in a store (joy 4% and trust 9% speak for themselves here). Analytically, however, statement 1 is much more interesting. What do we learn about the consumer who utters such a message?
We are struck by the high intensity of sadness and surprise, as it describes the state of the customer who has realized that he cannot fulfill his need. On top of that, there is a background of fear, probably caused by the fear of not having the favorite cheese in the future. The fear of shortages is accompanied by anticipation (will they deliver, won’t they?). Joy, on the other hand, is reflected in the fact of liking a certain kind of cheese, which is why it appears in the analysis in double digits.
Finally, we suggest looking at the table where we compare sentiment with specific emotions:
POLARYZACJA SENTYMENTU
POLARIZATION / SENTIMENT | |
| -
-
OCZEKIWANIE / EXPECTATIONS
-
|
| |
As you can see, assigning emotions from Plutchik’s model to only two types of sentiment can present difficulties when analyzing the data.
A few words of summary
“The pen is mightier than the sword”. – These were the words of Edward Bulwer-Lytton in his play “Richelieu”, and he was not wrong: words carry much more power than unreflective violence. No wonder scientists wanted to find out what lies behind words. What makes them appear at all and acquire some concrete meaning.
The first serious attempts to study sentiment began in the early 20th century, and had to do with studies of the polarization of public opinion. By the 1990s, there was already research into the subjectivity of textual content using computers, but the real explosion of research didn’t come until after 2004 – the year Google indexed 6 billion items, and Facebook debuted. There was research leading to the development of tools that could analyze sentiment. Who would have thought – almost 20 years have passed since then….
Today, content analytics technology has reached a new level: there is a shift away from simple, zero-one ratings of content to deeper analysis – emotion analysis. As a result, the scope of application of AI-based tools continues to grow. Today they can not only assess the mood of the consumer after a visit to the store; in this article we have used such examples because it is easy to show the principle of operation of analytical tools.
Tools for the analysis of emotions are already used in determining the behavior of investors on the stock exchange, prediction of stock prices or cryptocurrencies, they help in widely understood marketing activities, public relations (media monitoring) or customer service (e.g. analysis of communication with the chat-bot). There will be more and more applications, just as the technology itself will develop. Is it worth getting interested in? Definitely yes!
by Sentimenti Team | Aug 23, 2021 | SentiBrand
Getting to know consumer habits and understanding customer motivations is a good way to increase sales and grow your business. These goals can be achieved by analyzing emotions in the publications of Internet users, especially those related to your own brand. In this article we talk about the study of emotions in marketing and their impact on consumer behavior.
Emotions in marketing – are they worth exploring?
Emotions are responsible for certain body reactions and influence a person’s behavior. When analyzing emotions online, we use Robert Plutchik’s model, which distinguishes 8 basic emotions: joy, sadness, anger, disgust, fear, trust, surprise and anticipation. In expression, anger can lead to confrontation or aggression, joy enhances creativity and decision-making.
Sadness speaks of lack and manifests withdrawal, disgust rejects and stimulates flight, and fear also causes the urge to flee or attack. Trust speaks of closeness and relationship quality, surprise is an inducer of other emotions and feelings, and anticipation stimulates the body, preparing it for an upcoming event.
Since emotions are responsible for instinctive reactions, knowing their intensity can help predict consumer behaviour, improve customer service or monitor the brand image more effectively online. Knowing what evokes positive associations in customers, one can shape advertising, marketing and PR policy. Emotions are useful in creating characteristics of customers (persona) or for storytelling.
Researching customer emotions – advantages
Examining emotions in your own customers gives your company insight into consumer behavior. It will be useful e.g. when validating the effects of an advertising or communication campaign or in Content Marketing. It will help in creating tailored messages and more precise targeting of groups, more effective collection of leads, implementation of display campaigns, etc.
By examining the emotions of the recipients, the company may monitor its image and react in advance to the symptoms of crisis.
Emotion analysis versus sentiment analysis – the differences
The two terms are not used interchangeably; this is a mistake. Emotion analysis is the study of basic emotions – the unconscious, instinctual reactions of the brain and body to external or internal stimuli (e.g., thoughts, memories) and their effect on a person’s behavior. Meanwhile, sentiment analysis refers to reactions that are thought out and controlled by the subject: after becoming aware of the action of an emotion, he or she makes a decision that results in a mental attitude – sentiment.
Its study gives a result in the form of positive and negative sentiment, but we get an overall negative result without indicating the specific emotion and the behavior following it. With emotion analysis, we get the percentage score of each of the 8 emotions in the utterance, plus the sentiment analysis and the emotional arousal index.
They can check their own and their competitors’ emotionally charged phrases to determine how customers feel about their products or services, compare them with those of their competitors, and then make changes – to their offerings, their communication, or just their image.
Negative emotions: what can trigger them in marketing?
Negative emotions are conventionally called emotions whose perception is perceived as unpleasant. This perception gives the whole group of emotions a pejorative name, but the emotions themselves are warning signals of danger and are therefore not negative. In marketing or advertising emotions with such overtones may appear as a result of specific actions of brands or companies. What causes negative reactions of consumers?
For example, the use of the message, which strikes at the key values of the recipients of communication, breaks stereotypes, refers to unpopular views. Such slip-ups are the domain of multinationals that cannot predict the effects of actions in culturally distinct societies and do not take into account the mood of the target group. Another example is misalignment of the message with the requirements of the target group: it indicates misunderstanding of the group’s needs and will cause its frustration, which will translate into poor sales results and unfavorable comments on the Internet – and thus a scratch on the company/brand image.
Actions that cause negative emotions can also include overly intrusive PR and advertising, provocative actions (e.g. viral marketing), reprehensible practices towards employees, destruction of the environment, laboratory testing on animals, etc.
Negative customer emotions and their consequences
Customers most often talk about negative emotions through the company’s communication channels – social media, portals, e-mail. They comment under posts or create them themselves, review products and services, write opinions on forums, under articles, etc.
If the company’s message evokes negative emotions in them and these persist, their consequences will include negative WOMM (spreading unfavourable comments and opinions), brand switching (moving to the competition), brand detachment (severing relations with the brand), filing complaints, and even consumer boycotts, organizing protests or taking legal action.
How should a company respond to the negative emotions of its customers?
Negative emotions in your customers cannot be avoided, but you can minimize their effects. That’s why you should choose the analysis of emotions contained in the content. In social listening it will give you an up-to-date insight into the moods of your customers, in brand monitoring – the perception of the brand, it will also help to predict the behavior of consumers. And what to do when the symptoms of crisis appear? First of all, do not ignore them.
Negative moods will not subside on their own. Then accept the criticism and analyze the customer’s point of view – it is possible that the company’s policy was based on wrong assumptions. Finally – take concrete corrective actions, e.g. dialogue with the client, validation of communication or marketing activities, or improvement of the controversial service or product.
Interpretation of results:
The customer felt strongly surprised (62%) by the controversial statement of the maintenance department. She is angry about this (52%), but also feels anger about the careless finish of the apartment she bought.
The high level of surprise also relates to the individual faults that the customer mentions in her comment: the wrong way to suspend the ceiling, the faulty damp insulation around the chimney, the unprotected attic, the lack of a well-functioning but legally executed ventilation of the room with the fireplace or stove, the deficiencies in the electrical system and their repair that does not comply with building regulations, and finally the excessively high prices of additional services.
This state of affairs makes the customer feel sad (48%) and at the same time disgusted (loathing – 40%). The commentator does not know what else will happen to her in connection with the purchase (expectation – 35%), is afraid that things will not turn out well (fear – 36%), but still has hope for a positive outcome (expectation – 35%, trust – 21%). Finally, there is the indicator of joy (22%).
In the present case, it refers to ironic comments towards the developer, who is satisfied with his actions. We also have results for sentiment and emotional agitation – positive – 21%, negative – 45%, emotional agitation – 67%. This last element indicates that the commenter is highly agitated and inclined to take action.
What actions from such a client should be expected? Her emotional state, accompanying negative sentiment, and high arousal rate will likely push the commenter to post negative comments (negative WOMM) on forums, social media, and anywhere else she sees requests for feedback about this particular developer. In all likelihood, she will advertise any faults that arise with the developer, and possibly – pursue legal action.
Author: Igor Starczak. The publication also appeared in the quarterly “Developer & Marketing” (No. 3 / 2021).
by Sentimenti Team | Aug 5, 2021 | Conferences, Market research, Scientific publications
Place of publication:
- Conference: Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing
Title:
Personal Bias in Prediction of Emotions Elicited by Textual Opinions
Authors:
Jan Kocoń, Piotr Miłkowski, Damian Grimling, Marcin Gruza, Kamil Kanclerz, Przemysław Kazienko
Abstract:
Analysis of emotions elicited by opinions, comments, or articles commonly exploits annotated corpora, in which the labels assigned to documents average the views of all annotators, or represent a majority decision. The models trained on such data are effective at identifying the general views of the population. However, their usefulness for predicting the emotions evoked by the textual content in a particular individual is limited. In this paper, we present a study performed on a dataset containing 7,000 opinions, each annotated by about 50 people with two dimensions: valence, arousal, and with intensity of eight emotions from Plutchik’s model. Our study showed that individual responses often significantly differed from the mean. Therefore, we proposed a novel measure to estimate this effect – Personal Emotional Bias (PEB). We also developed a new BERT-based transformer architecture to predict emotions from an individual human perspective. We found PEB a major factor for improving the quality of personalized reasoning. Both the method and measure may boost the quality of content recommendation systems and personalized solutions that protect users from hate speech or unwanted content, which are highly subjective in nature.
Link: ResearchGate
