Sentiment analysis

<h3>2.1. Sentiment analysis</h3>

Sentiment analysis is a scientific field that examines and analyzes the subjective content of textual data from the conversational content of social networks. It mainly focuses on the analysis of the polarity of opinions, attitudes, and emotions of people to determine their satisfaction or dissatisfaction with the object of discussion. It is a challenge for projects in the field of natural language processing, computational linguistics, and text mining. Sentiment analysis includes the following sub-problems (

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#ref23">Machová et al., 2020a</a>):

•&nbsp;

Subjectivity detection&nbsp;aims to distinguish subjective from neutral terms, phrases, sentences or comments and is frequently used as an initial step in polarity and intensity of polarity recognition, to separate subjective information from objective ones.

•&nbsp;

Polarity classification&nbsp;attempts to classify texts into positive, negative, or neutral classes. It forms the basis for determining the polarity of the text as a whole.

•&nbsp;

Intensity classification&nbsp;goes a step further and attempts to identify the different degrees of positivity and negativity, e.g., strongly negative, negative, fair, positive, and strongly positive. Also, special words “intensifiers” are used for intensity classification. They can increase or decrease the intensity of polarity of connected words, e.g., surprisingly good, highly qualitative.

•&nbsp;

Opinion spam&nbsp;is another problem inhibiting accurate sentiment analysis. Spam distorts product quality evaluation and precision of the polarity recognition of an opinion.

•&nbsp;

Negations processing&nbsp;is used when negation before a word changes the polarity of a connected word. The most used negation processing methods are the switch and the shift negation.

•&nbsp;

Emotion detection&nbsp;seeks to identify if a text expresses any type of emotion or not. Also, a problem of identification of the polarity of detected emotion is often necessary.

We have focused on emotion detection and on the possibilities of using it in social and psychological domains.

<h3>2.2. Emotion analysis</h3>

Extracting context from text is one of the most remarkable acquisitions obtained with natural language processing (NLP). A few years ago, context extraction was supposed to detect the polarity of sentiment from text, then the world took a step forward to detect sentiment in the form of emotions. These two concepts are very different. Sentiment can be positive, negative, or neutral, while emotions are more refined categories between positive and negative. Positive sentiment can be attributed to a happy, joyful, excited, and even funny emotion. Similarly, anger, disgust and sad emotions cause the sentiment to be negative. Several years ago, many machine learning algorithms were used in experiments to be training emotion detection models, but also to use a lexical approach for emotion recognition based on lexicons as lists of emotional words typical for specific emotions. However, all these approaches are slowly becoming obsolete due to the new trends in the deep learning detection models, which can do a very accurate automatic analysis of emotions from a text. The most well- known and successful models being CNNs and recurrent neural networks (RNN), particularly LSTM.

A valuable survey of approaches of emotion recognition was made by&nbsp;

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#ref8">Ekman (2016)</a>. Three major directions in emotions recognition are: categorical/discrete, dimensional, and appraisals-based approaches (

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#ref14">Grandjean et al., 2008</a>):

• Basic emotion model: the categorical approach - claims there are a small number of basic emotions that are hard-wired in our brain and recognized across the world. Each affective state is classified into a single category.

• Dimensional feeling model: the dimensional approach - based on Wundt’s proposal (

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#ref9">Feldman-Barrett, 2018</a>) that feelings (which he distinguishes from emotions) can be described as pleasantness–unpleasantness, excitement–inhibition and tension–relaxation. An example is Plutchik’s wheel of emotions illustrated in&nbsp;

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#fig1">Figure 1</a>. In Plutchik’s wheel of emotions, primary, secondary, and tertiary dyads are presented. Each dyad describes the distance between two emotions (hence the term dyads) as follows: the primary dyad combines emotions next to each other (i.e., Joy – Trust in&nbsp;

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#fig1">Figure 1</a>); the secondary dyad combines two emotions when another emotion between them is skipped (i.e., Joy – Fear with skipped Trust in&nbsp;

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#fig1">Figure 1</a>); the tertiary combines two emotions when two other emotions between them are skipped (i.e., Joy – Surprise with skipped Trust and Fear in&nbsp;

<a href="https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2023.1190326/full#fig1">Figure 1</a>).

• Componential appraisal models: proposes that emotions are extracted from our appraisals (i.e., our evaluations, interpretations, and explanations) of events. These appraisals lead to different specific reactions of different people. It defines emotions as a balanced reaction to events, agents, and objects, and considers balanced reactions to differentiate between emotions and non-emotions. This approach is very suitable for affect sensing from the text.



&nbsp;

Figure

<img src="https://firebasestorage.googleapis.com/v0/b/imentiv-assets/o/assets%2F83227d1a-ded8-401e-b1fd-666597b2394afpsyg-14-1190326-g001.jpg?alt=media&amp;token=5a7c4fb4-111b-4e74-a4c4-712db238aad4" alt="Image" style="height:1414px;width:1417px;">