helpfn

Are They Different? Affect, Feeling, Emotion, Sentiment, and Opinion Detection in Text Myriam Munezero, Calkin Suero Montero, Member, IEEE, Erkki Sutinen, and John Pajunen

Abstract—A major limitation in the automatic detection of affect, feelings, emotions, sentiments, and opinions in text is the lack of

proper differentiation between these subjective terms and understanding of how they relate to one another. This lack of differentiation

not only leads to inconsistency in terminology usage but also makes the subtleties and nuances expressed by the five terms difficult to

understand, resulting in subpar detection of the terms in text. In light of such limitation, this paper clarifies the differences between

these five subjective terms and reveals significant concepts to the computational linguistics community for their effective detection and

processing in text.

Index Terms—Affect sensing and analysis, computation models of emotion

Ç

1 INTRODUCTION

THE ability to accurately identify and reflect “what is sub-jectively written” is valuable. This ability influences real-world text analysis applications such as targeting mar- keting campaigns to users with specific likes and dislikes or identifying online antisocial behavior, for example, cyber- bullying. Thus, being able to understand and detect the sub- jective nuances in text is valuable for its actual applications.

Affect, feeling, emotion, sentiment, and opinion are terms relating to human subjectivity. Because subjectivity is a feature of a person’s mind, it “is not open to objective observation or verification” [1]. Subjective experiences are created from the subject perspective, and they reflect the subject’s desires, beliefs and feelings. Thus, the detection of subjectivity is a difficult natural language processing (NLP) task yet intriguing as a research endeavour.

Current text-based subjectivity detection efforts, how- ever, are limited in their detection abilities. For instance, most of the sentiment analysis (SA) and opinion mining (OM) studies have been focused on detecting text polarity (positive, negative, and neutral) [2]. However, sentiments and opinions are more complex than just having polarity. Moreover, subjectivity terms indicating affects, feelings, emotions, sentiments, and opinions in text are often under- stood similarly and used interchangeably in NLP research without sufficient differentiation between them. This lack of consistency in terminology usage leads to poor apprehen- sion and confusion about what concepts or features should be involved in the text analysis, as also noted by Hovy [3], Balahur et al. [4] and Alm [5].

Thus, to overcome the outlined limitations, we went a step further and clearly established a distinction between the different branches of the subjectivity terms. Although there are similarities, subjectivity detection advances toward more fine-grained analysis of human subjectivity, compelling the subjectivity terms’ subtle differences to play an important role in increasing the effectiveness of NLP detection methods.

To reach their full potential, the NLP analysis must be able to capture the subtle differences that reflect personal, cultural, and societal signals within the subjectivity terms. Therefore, it is fundamental to establish concise and clear terminologies. In this paper, we aimed to shed light on the differences among subjectivity terms and put forward the potential that these differences provide to advancing the NLP research.

2 SUBJECTIVITY TERMS

Standard dictionaries are an initial resource with which to analyze the meanings of words. Table 1 provides the defini- tions of affect, feeling, emotion, sentiment, and opinion along with their synonyms, based on the Merriam-Webster Online Dictionary [6].

From the definitions and the synonyms given in Table 1, the source of confusion in meaning and usage is shown by the repetition of the shared synonyms. The confusion is further compounded by the common everyday usage of the terms [7]. For example, when a person says “my sentiments exactly!” it often has the same meaning as “my feelings exactly!” Thus, the subtle differences as they are used in research literature will be revealed only by diving deeper into the definitions of the above terms. Moreover, it will also be possible to model or detect the subjective terms as they exist by outlining the differences.

3 SUBJECTIVITY TERMS IN THE PSYCHOLOGICAL LITERATURE

3.1 Affects

Among all of the subjectivity terms, affect seems to be the most abstract and difficult to be fully realized in language

The beginning of wisdom is the definition of terms. —Socrates

� M. D. Munezero, C. Suero Montero, and E. Sutinen is with the School of Computing, University of Eastern Finland, PL 111, 80101 Joensuu, Finland. E-mail: {myriam.munezero, calkin.montero, erkki.sutinen}@uef.fi.

� J. Pajunen is with the Department of Philosophy, University of Jyv€askyl€a, Seminaarinkatu 15, 40014, Jyv€askyl€a, Finland. E-mail: [email protected].

Manuscript received 23 June 2013; revised 13 Feb. 2014; accepted 2 Apr. 2014. Date of publication 13 Apr. 2014; date of current version 23 July 2014. Recommended for acceptance by R. Mihalcea. For information on obtaining reprints of this article, please send e-mail to: [email protected], and reference the Digital Object Identifier below. Digital Object Identifier no. 10.1109/TAFFC.2014.2317187

IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. 5, NO. 2, APRIL-JUNE 2014 101

1949-3045 � 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

as it is always prior to or outside of consciousness [8]. Affect is the “body’s way of preparing itself for action in a given circumstance by adding a quantitative dimension of inten- sity to the quality of an experience” [9]. Affect within the psychological literature has been thought of as the umbrella term; it subsumes emotions, feelings, and sentiments [10]. Batson et al. [11] found affect to be the most general of the terms, describing it as more phylogenetically and ontologe- netically primitive than emotions and further explaining that “affect is present in the yelp of a dog and in the coo or cry of an infant.” Shouse [9] defined affect as a “non-con- scious experience of intensity: as a moment of unformed and unstructured potential.” He adds that affect “plays an important role in determining the relationship between our bodies, our environment, and others.” Thus, affect can be defined as positive and negative evaluations of an object, behavior, or idea with intensity and activity dimensions [12]. Zajonc [13] similarly agreed that affect reveals prefer- ences; that is, it informs the organism experiencing it about those states of affairs that it values more than others. Change from a less valued to a more valued state is

associated with positive affect; change from a more valued to a less valued state is associated with negative affect. The intensity of the affect then reveals the magnitude of the value preference [11].

3.2 Feelings

Shouse [9] defined a feeling as a “sensation that has been checked against previous experiences and labeled.” It is per- sonal and biographical because every person has a distinct set of previous sensations from which to draw from when interpreting and labelling their feelings.” Shouse made an interesting distinction between feelings and affect in infants, in that infants do not experience feelings because they lack both language and biography. He said that what infants experience is affect. Moreover, Thoits [12] stated that feel- ings are the experience of physical drive states such as hun- ger, pain, fatigue, and emotional states. They are part of our “psychological and physical baggage” [10]. Friedenberg and Silverman [14] added that feelings “correspond to the subjective experience of an emotion.” Wierzbicka [15] con- firmed this by stating that “feelings are subjective, and they appear to be universally thought of as related in some cases to what is happening in the body; but they are also often thought of as based on certain recurrent thoughts—cogni- tive scenarios shaped by the particular culture.” Wierzbicka [15] additionally pointed out that the concept of feelings is universal—“all languages appear to have a word for the concept ‘feel,’ we can assume that this concept is an integral part of the universal folk model of a person, that is, that in all cultures people attribute feels to other people as well as to themselves. Furthermore, evidence suggests that in all cultures people distinguish linguistically (and in particular, lexically) between different kinds of feelings.” In this work, we do not consider feelings that are not thought related, such as a “feeling of hunger” or a “feeling of cold,” as it is not comparable with the other affect terms. It is clear that one will not talk about an affect, an emotion, a sentiment or an opinion of hunger, cold, and so on [15].

3.3 Emotions

Definitions of emotions in the literature have been many and varied. Kleinginna and Kleinginna [16] performed a compilation of 92 definitions and nine skeptical statements from a variety of emotion literature sources. On the basis of their compilation, they suggested a formal definition of emotion as “a complex set of interactions among subjective and objective factors, mediated by neural and hormonal sys- tems, which can a) give rise to affective experiences such as feelings of arousal, pleasure and displeasure; b) generate cognitive processes such as emotionally relevant perceptual affect, appraisals, labeling processes; c) active widespread physiological adjustments to the arousing conditions; and d) lead to behavior that is often, but not always expressive, goal-directed and adaptive.”

In his work on emotion psychology, Scherer [7] blamed the confusion in the definition of emotions to the failure to clearly distinguish between the different classes of phenom- ena. Scherer advised that the general affective valence or preference should not be treated in the same manner as emotional episodes, nor should it be to the more enduring

TABLE 1 Definitions Provided by Merriam-Webster Online Dictionary [6]

102 IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. 5, NO. 2, APRIL-JUNE 2014

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

affective states such as attitudes. However, emotions as a type of affective state do have valence and intensity [11]. Scherer settled on a working definition of emotion as “episodes of coordinated changes in several components (including at least neurophysiological activation, motor expression and subjective feeling but possibly also action tendencies and cognitive processes) in response to external or internal events of major significance to the organism.” Thoits [12] defined emotions as culturally determined types of feelings or affect. Shouse [9] rephrased to say that emo- tions are the projections or display of a feeling, and unlike feelings, the display of emotions can be genuine or feigned. Shouse adds that we “broadcast emotion to the world; sometimes that broadcast is an expression of our internal state and other times it is contrived in order to fulfill social expectation.” Dolan [17] believed that from a psychological perspective, emotions have three characteristics: “first unlike thoughts, emotions are embodied, that is we experi- ence emotions not just as mental events that are in our heads but also as full-body experiences. Secondly, unlike cogni- tion, emotions are harder to control. We may find that it is easier to change our thoughts than our emotions. Thirdly, emotions seem less encapsulated than thoughts; that is they seem to have a more global impact on our behavior.” Batson et al. [11] further described emotions as reflections of the “existence of a specific goal or of perceived change in one’s relationship to a specific goal in the present.” Goals, as Batson pointed out, reflect values, and what is valued depends on an individual’s current situation, which draws from personal experiences, culture, and society.

Friedenberg and Silverman [14] simply considered emo- tion as “brief brain and body episode that facilitates a response to a significant event.” Continuing from Shouse’s example, infants display emotions although they do not have the biography or the language skills to experience feel- ings. The emotions of an infant are direct expressions of affect. Thus, according to Shouse [9], emotions are the expression of affect and/or feelings

3.4 Sentiments

Sentiments are defined by Cattell [18] as “an acquired and relatively permanent major neuropsychic disposition to react emotionally, cognitively, and conatively toward a cer- tain object (or situation) in a certain stable fashion, with awareness of the object and the manner of reacting.” Gor- don [19] has a similar definition; sentiments are “socially constructed patterns of sensations, expressive gestures, and cultural meanings organized around a relationship to a social object, usually another person (. . .) or group such as a family.” Similarly, Murray and Morgan [20] defined senti- ment as “a more or less enduring disposition (predilection or readiness) in a personality to respond with a positive or negative affect to a specified entity.”

Cattell [18] further pointed out that the “functioning of sentiments is accompanied in the individual by interest and a sense, on his part, of ‘values’ in the object.” Thus, the object has come to possess more or less an enduring power to “evoke relatively intense and frequent reactions, positive or negative in a person” [20]. Broad [21] explained that a sentiment is formed when a certain object is constantly per- ceived or thought of by a person and, over time, the person

creates a dispositional idea toward the object. This disposi- tional idea has corresponding emotional tendencies that are evoked whenever the person perceives and thinks about the object or any symbols related to the object.

Examples of sentiments include romantic love, parental love, loyalty, friendship, patriotism, hate, as well as more transient, acute emotional responses, to social losses (sor- row, envy) and gains (pride, gratitude) [12].

3.5 Opinions

Opinions in research have been synonymous to what a per- son is thinking. An opinion is the idea or knowledge one has about something. In Pang and Lee’s work [2], an opin- ion “implies a conclusion thought out yet open to dispute”; that is, experts on a topic can have very different opinions, which they might change in light of new evidence. Psycho- analysts describe opinions as rationalizations, explanations, and justifications [22].

Young [22] further defined an opinion as a judgement based on grounds insufficient to produce certainty. Thus, opinions leave room for error in thought. Young [22] charac- terizes opinions as provisional convictions, views held as probable, or a belief or disbelief in some idea or form of con- duct. Beliefs, as Fishbein and Ajzen [23] described, are the “the subjects perception that an object or person has certain characteristics, qualities, or attributes, or is related to some other concept, object, or person.”

According to Kim and Hovy [24], an opinion consists of the following four parts: topic, opinion holder, claim, and sentiment. That is, for each opinion, there is a holder who believes a claim about a topic and then associates a positive, negative, or neutral (neutral here does not mean absence, e.g., “The winter has arrived. It is not good or bad, just saying”) sentiment with the claim. Fig. 1 illustrates how the four aspects can appear in a sentence. Kim and Hovy [24] further pointed out that an opinion can be subjective without implying a sentiment. This was also acknowledged by Sokolova and Bobicev [25], who defined opinions as “subjective statements that may or may not imply emotions.”

Liu [26] rather used a mathematical representation of an opinion as a quintuple <o; f; so; h; t> , where o is an object; f is a feature of the object o; so is the orientation or polarity of the opinion on feature f of object o (positive, negative or neutral); h is an opinion holder; t is the time when the opin- ion is expressed.

4 DIFFERENTIATING FACTORS OF SUBJECTIVITY TERMS

4.1 Affects, Feelings, and Emotions

Massumi [27], in his “Notes on the Translation and Acknowledgement,” saw affect as “prepersonal”; that is,

Fig. 1. Example of an opinion (based on Kim and Hovy [24]).

MUNEZERO ET AL.: ARE THEY DIFFERENT? AFFECT, FEELING, EMOTION, SENTIMENT, AND OPINION DETECTION IN TEXT 103

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

affect exists outside of consciousness before personal self- awareness develops. Shouse [9] further expands this idea by giving the example of affective expression in infants when the expression of feelings is not yet developed. Fur- thermore, Matthis [28] defined affect as an “affective matrix linked to a predisposition of the bodily systems to react in a certain way to internal or external stimuli” and also as a “higher-order category under which both feelings and emo- tions fall.”

In contrast, feelings are seen more tangibly as a “broad category of person-centered psycho-physiological sensations” [29] in [30], as “affective phenomena to which we have direct conscious access” [28], and as labeled sen- sations, “personal and biographical” [9].

Emotions are seen as preconscious phenomena; that is, they are capable of becoming conscious upon recall. Emo- tions are defined based on the expression of feelings as a “subset of particularly visible and identifiable feelings” [29] in [30]; as “affective manifestations to which we do not have direct conscious access, but which can be inferred from behavioural clues” [28]; and as a “social . . . projection/dis- play of a feeling” [9]. Furthermore, Wierzbicka [15] noted the influence of culture on emotions and stated that although “feeling is universal and can be safely used in the investigation of human experience and human nature, the concept of emotion is culture-bound and cannot be similarly relied on.” Therefore, the way in which emotions are expressed is more often than not a result of the culture/ society one lives in.

We, thus, find differentiating factors between affect, feel- ings, and emotions, as follows:

� Affect is, in fact, a predecessor to feelings and emotions.

� Feelings are person-centered, conscious phenomena. � Emotions are preconscious social expressions of feel-

ings and affect influenced by culture.

4.2 Sentiments and Emotions

The concepts of emotions and sentiments have often been used interchangeably. This is mostly because both emotions and sentiments refer to “experiences that result from the combined influences of the biological, the cog- nitive, and the social” [31]. However, sentiments are dif- ferentiated from emotions by the duration in which they are experienced.

Although emotions are brief episodes of brain, auto- nomic, and behavioral changes [32], sentiments have been found to form and be held for a longer period. Sen- timents, in addition, are more stable and dispositional than emotions [33]. BenZe’ev [33] further explained that a person may have a sentiment of, for example, love when he or she does not actually have an occurring state of love.

Moreover, sentiments are formed and directed toward an object1, whereas emotions are not always targeted toward an object. For instance, one can wake up feeling depressed or happy for no apparent reason [34].

4.3 Sentiments and Opinions

Sentiments and opinions have also been used interchange- ably, perhaps because most NLP research on opinions has focused on detecting their subjective part, which has been referred to as sentiment (see Kim and Hovy [24]). How- ever, opinions are personal interpretations of information formed in the mind and not necessarily constrained by social expectations, whereas sentiments are generally socially constructed. In addition, opinions are prominently based on objective and/or subjective probabilities of infor- mation about a topic, whereas sentiments are prompted by emotions.

Cattell [18] distinguished between sentiments and opin- ions in his analysis by pointing out that an individual pos- sesses a sentiment and experiences a complex secondary or derived emotion, whereas an opinion is not possessed but is an expression of personal interpretation of information. Sokolova and Lapalme [35] acknowledged that sentiments should not be treated as identical to opinions because opin- ions can be expressed about a fact of matter, in a descriptive and assessing way, without the expressions being emotion- ally charged [36].

Hence, we find differentiating factors between senti- ments and opinions, as follows:

� Sentiments are partly social constructs of emotions that develop over time and are enduring.

� Opinions are personal interpretations of information that may or may not be emotionally charged.

Fig. 2 serves to illustrate the differentiating factors of the subjective terms.

5 ADVANCING THE DETECTION OF SUBJECTIVITY TERMS IN TEXT

From the definitions of the terms, we observe that the cur- rent NLP approaches will benefit from a deeper analysis of

Fig. 2. Differentiating factors between affect, feelings, emotions, senti- ments and opinions.

1. An object refers to a person, a thing, a condition, a place or an event at which a mental state is directed [34].

104 IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. 5, NO. 2, APRIL-JUNE 2014

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

affect, feelings, emotions, sentiments and opinions in text. In our attempt to expand the detection approaches of these terms in text, we put forward how the differentiating factors we identified in Section 4 should be realized and applied by building on the current NLP advances.

5.1 Detecting Affect

Affective computing (AC) as introduced by Picard [37], is a broader research field concerned with giving computers the ability to recognize, express, and, in some cases, “have” emotions and other related affective phenomena. AC explores the recognition of affect in faces, voice, pos- ture, or other mannerisms and behaviors as a multimodal research field.

In Picard’s [37] work on AC, the terms emotion and affect are used interchangeably. This is quite understandable because, as we have seen, emotions are in fact expressions of affect. However, as already discussed, affect is difficult to realize in language. Conversely, when we perceive the world around, we primitively have an affective reaction (good or bad) [13] to the objects in it. That is, we perceive objects to be beautiful, ugly, awful, and so on. These affec- tive reactions are the conscious representation of affect, which in language is expressed through feelings and emotions.

5.2 Detecting Feelings and Emotions

Unlike affect, emotions and feelings have been affirmed by several researchers as being expressible in language ([38], [39], [40], [41]). However, because of its complexity, emo- tions are not always fully accessible in written language. In this section, we look at both emotions and feelings as emo- tions are expressions of feelings (see Fig. 2).

From the definitions of emotion we considered in Section 2, particularly in the summary of Kleinginna and Kleinginna’s [16] study, we find that there seems to be a rea- sonable agreement over the following relevant factors that determine an emotion:

a) Appraisal (cognition). b) Physiological reactions of the body, such as

increased heartbeat and sweating. c) Feeling. d) Expressive display, such as facial expression and

bodily expression. e) Readiness to behave in a particular way (also called

action tendencies by Frijda [42]). These five factors of emotion are not independent of

one another, and in fact, changes in one factor can lead directly to the corresponding changes in others. How- ever, it is not certain which factor precedes which, or if all the factors are necessary during an emotional experi- ence [43]. Nevertheless, Russell and Barrett [34] pointed out that the fewer factors assessed, the less convincing the results. Furthermore, these factors are also influenced by culture and societal norms. Culture and society are identified here as fundamental parts of emotion expres- sion because emotions are culturally/socially constrained expressions of affect. Hence, social processes and cultural norms play a significant role in specifying how we express emotions [44].

However, Mesquita and Leu [45], in their cultural psy- chology of emotion research, acknowledged the possibility of some aspects of emotions, for example, facial expressions and other physiological reactions, to be hardwired and com- mon across cultures. Nevertheless, they also put forward that there are cultural differences in each of the above men- tioned five factors of emotion. As an example, Mesquita and Leu [45] found insults were appraised as an honor violation and were more likely to result in anger and aggressive responses in the south than in the north of the United States where honor culture is not strong.

The identification and assessment of some or all of the factors would provide a better analysis of emotions in text. Conati [46], for example, combined evidence from appraisals that produce emotions and bodily expressions associated with emotion expression for his analysis of emotional experiences during learning. However, there are limitations incurred with text environments in that they cannot deliver a full emotional experience by them- selves [47]. That is, it is not possible to access or measure directly the physiological changes in the body of the writer through text, for instance. Regardless, it is possi- ble for all the five factors to be expressed in language to some extent.

5.2.1 A Scenario

To illustrate how the emotional experience factors could be lexically encoded in text to communicate an emotion that could be labeled as anger,2 let us consider a hypothetical scenario where a friend communicates an event to another friend.

“Hi John, remember my friend Luke? Can you believe that he punched me on Friday at John’s party! I can’t believe my friend would hit me, he was acting crazy. Never felt so insulted in my life. How could he do that!!! Just made me lose my temper. I wanted to just hit him right back too, could feel the adrenaline rushing in. Started swearing at him and pointing a certain finger to him. You should have been there. Wanted to kick the guy’s a��, but I thought it was not was not worth it you know.”

From the story we can observe the following emotion factors:

- Appraisal—“I can’t believe my friend hit me, he was crazy”

� This expression is the appraisal of the eliciting event, which is assessed as negative and unde- sirable to one’s social interaction and well- being. Appraisals are evident in text through the use of evaluative language (for identifying expressions of appraisal in text, see Read et al. [49]). Appraisals in text can be used as evidence of an emotion.

- Physiological reaction—“adrenaline rushing in”

� Certain phrases can be used to express the inner physiological arousal. References to heart rate

2. Frijda [48] explained that “anger, in English, tends to focus on either the appraisal that someone else was doing something unpleasant which was not felt to be in line with accepted rules or on the impulse to oppose and counteragress.”

MUNEZERO ET AL.: ARE THEY DIFFERENT? AFFECT, FEELING, EMOTION, SENTIMENT, AND OPINION DETECTION IN TEXT 105

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

(my heart skipped a beat), temperature (getting cold feet), muscles tightening (knots or butterflies in the stomach), breathing (finding it hard to breath), and so on.

- A (negative) feeling—“made me loose my temper”

� Feelings are made clear using descriptive or expressive language. They can be explicitly stated with emotion labels such as “angry” or “furious”, or they could be expressed implicitly “made me lose my temper”.

- Expressive display—“swearing at him and pointing a certain finger to him”

� Expressive behavior such as facial expressions can be found in text in the use of emblems, such as emoticons (e.g., ) and with references to gestures and postures (e.g., shaking one’s head in disap- proval, fiddling, clearing of throat, quivering, grin stretching from ear to ear).3

- Readiness to behave in a particular way—“wanted to just hit him right back”

� Behaviors are expressed using verbs (e.g., “hit”). As can be seen in this scenario, beyond the use of emo-

tion-bearing words, an author is equipped with a variety of linguistic markers and their combinations to convey an emotion [50]. These markers include stylistic features such as punctuations and capitalization [51]. For emotion detec- tion, it is particularly important to go beyond word-level analysis, as single words by themselves are not able to carry all the emotional information: emotions are often hidden behind combinations of words [52].

5.2.2 Labeling Emotions as Shaped by Culture

and Society

Of interest to NLP is being able to tell which emotion is expressed in the text. Predominantly, research on detecting emotions from text has focused on capturing emotion words based on three emotion models, that is, categories of basic emotions [53], emotion dimensions [54], or cognitive- appraisal categories [55]. These models differ greatly with respect to the number of emotions they explain as well as the principles that they evoked [7]. Unfortunately, as of date, no general model of emotions has yet been agreed on [56]. It is not the aim of this paper to argue on the correct model to use in labeling emotions in text because they are all founded on different psychological theories of emotion, and they all fit specific purposes according to the goals and purpose of a particular research in various fields.

In computational linguistics, attempts at detecting emo- tions have been made using lexical resources such as Word- NetAffect [57] to mark the presence of emotion-bearing words in text or Affective Norms for English Words (ANEW) [58] to mark words with normative emotional rat- ings. Also, commonsense knowledge bases such as Emoti- Net [59] have been used to capture emotions in texts that do not have emotion-bearing words.

Central to this paper is the call to extend the NLP approaches to include the analysis of words, sentences, or documents on a cultural level. This is because naturally the choice of words and their intended meaning are per- sonally, contextually, culturally, and socially dependent. That is, a word, a phrase, or a sentence might have one emotional value when considered in isolation, but when evaluated on a cultural or societal context, the emotional value might change. An example of this can be observed in the following sentence: “That ride is wicked.” By defi- nition of the word “wicked”, the sentence has an unpleas- ant connotation in some communities; however, the sentence would be associated with pleasantness in some communities such as the United Kingdom. Similarly, the word “blue” was observed by Das [60] in their work as being positive in many countries but was seen as negative around the Middle Eastern region. Das [60] illustrated the possibility of encoding cultural information in lexical resources to create a more robust resource for emotion detection.

Thus, regardless of the model chosen to represent the emotions (e.g., Ekman’s six basic emotions [53] or Russell’s dimensions [54]) or the approach used, the researcher should also evaluate a writer’s emotional expe- rience on a cultural, societal level. This could be done primitively by accessing the location information of the analyzed text. In the case of online sources, for instance, Facebook [61], it is possible to access the location4 or nationality information of a user whenever that informa- tion is made public. Hence, investigations of automatic detection of emotions have the means to deploy a wider ethnographic inquiry into the words used to express emo- tions. Fig. 3 summarizes the proposed structure for analy- sis for emotions from text.5

Fig. 3. Schematic structure of an emotion.

3. We must keep in mind, however, that the same expressive behav- ior might indicate a different emotional experience, for example, crying when feeling happy or sad.

4. Here it is supposed that the location where the text is generated corresponds to the physical world where the writer is and, hence, may provide clues on his country or society.

5. However, one must also be careful and keep in mind that what is expressed might not be what is experienced. For example someone say- ing “I hate you” might just be saying it as a joke and not be really experiencing the emotion “hate”.

106 IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. 5, NO. 2, APRIL-JUNE 2014

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

5.3 Detecting Sentiments

Expressions of Sentiment, by default, have almost been defined in NLP as a positive or negative expression. How- ever, from our analysis, they are a more complex phenome- non. Sentiments involve emotional dispositions formed toward an object, and unlike emotions which are brief, sen- timents about an object are enduring. Thus, labeling brief emotional episodes or affective qualities as sentiments is a misrepresentation.

5.3.1 Sentiment Endurance

French [62] pointed out that increasing experience with an object “brings about differentiation of aspects of both posi- tive and negative valence and in this turn leads to conscious ambitendency.” Shelly [63] also acknowledged that increased experience with an object leads to enduring pat- terns conceptualized as liking and disliking toward the object. These two broad categories of like and dislike are sufficient enough to include both positive and negative emotions that may arise whenever the liked or disliked object comes to mind.

Approaching sentiments as enduring is useful for research in domains such as terrorism or antisocial behavior, as they indicate dispositions that are not about to change unless something intervenes. It could also provide better informed reviews; that is, it could assist in judging the rele- vancy or impact of reviews. A bad review of the movie Sky Fall from a person who generally enjoys and likes action movies, for instance, would probably have more relevance for recommendation systems than a bad review from some- one who does not have a liking for action movies. Fig. 4 illustrates our formulation of the structure of a sentiment.

However, the field dedicated to detecting sentiments, SA, has not treated sentiments as enduring. Realistically, senti- ments can only be inferred on the basis of the regular occur- rence of emotional responses to an object [62]. As an experiment, we made efforts to infer enduring sentiments based on emotions. The method was practical enough to allow us to collect freely available user data from Twitter over a period ranging from three months to one year, depending on the user. The user’s tweets were gathered and filtered according to an object (e.g., person, place, and situation). With the filtered tweets, we analyzed the emo- tions along Plutchik’s eight basic emotions [64]: joy trust, anticipation, surprise, anger, fear, disgust, and sadness.

These emotions were then mapped into two broad cate- gories: like and dislike. Finally, a total score of like and dislike was calculated over the period of the tweets as a function of the emotions present in the texts. This approach allowed us to visualize the user’s regularly occurring dispositions toward the object in a certain period6 [65]. The advantage of approaching sentiments in this manner is that the emotional disposition formed toward an object can be observed along- side the comments made by a user. Such information is use- ful, for instance, in computing the loyalty of customers toward a product. For example, looking at tweets made by a user, we are able to see that the person dislikes the iPhone5 but likes the new iPad and, overall the brand Apple.

5.4 Detecting Opinions

In computational linguistics and NLP, the field dedicated to the research and detection of opinions in text is called opin- ion mining. As mentioned previously, the terms OM and SA have often been used interchangeably, but as illustrated in Fig. 2, opinions can be based on the personal interpretation of information but may not carry a sentiment or an emo- tion.7 Thus, with the definition of sentiment as discussed in Sections 2 and 3, it would be misleading for NLP research- ers to treat the concepts as the same.

From the definitions of opinions, we find that an opinion includes at least three parts: topic, opinion holder, and claim. In addition, an opinion may include the following: (1) an emotional and sentimental expression, (2) feature(s) of the topic, and (3) time (the time that an opinion is made).

Particularly of interest to OM is to determine whether the expressed opinions are positive, negative, or neutral. An opinion can be categorized as neutral when it is stated as a matter of fact without any positive or negative evalu- ation, for example, “Bin Laden is hiding in Pakistan”.8

Moreover, if an opinion includes an emotion or sentiment expression, then the expressed emotions can be used to determine the polarity of the opinion as negative or posi- tive. In the case that no emotion or sentiment is attached, the polarity of an opinion can still be determined in two ways, as follows:

1. By analyzing if there is an affective reaction asso- ciated to the topic or feature (“I found this book boring” vs. “this movie was directed beautifully”).

Fig. 4. Schematic structure of a sentiment.

6. We acknowledge that outside of twitter and diaries, it is not always possible to find one person’s thoughts and emotional disposi- tions toward an object over a long period for analysis.

7. Note that we used a sentiment or an emotion. This is because a person might form an opinion based on a brief emotional experience (“I loved that movie”, just after coming from the cinema) or based on a disposition they have had (“Action movies are the best”, based on the liking disposition a person has developed for action movies). Unfortu- nately, in many domains where OM is applied, for example, product or customer reviews, the brief versus enduring distinction is not easy to obtain.

8. Sayeed [66] pointed out that the polarity of an opinion is depen- dent on the perspective of the reader of the text. For example, to a casual reader, the sentence “Bin Laden is hiding in Pakistan” might be perceived as a neutral sentence, but to an interested reader, it might represent a positive or a negative statement depending on whether knowing the location of Bin Laden is a good thing or not. However, we restricted our analysis of the subjective terms from the perspective of the author.

MUNEZERO ET AL.: ARE THEY DIFFERENT? AFFECT, FEELING, EMOTION, SENTIMENT, AND OPINION DETECTION IN TEXT 107

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

Research on the polarity of opinions based on the affectively charged reactions on a topic has been performed using current NLP approaches, for instance, using of affective keywords from exist- ing knowledge bases (WordNet [67]) or from pur- pose built lexicons based on domain-dependent corpora ([68], [69], [70], [71]).

2. By analyzing the words people choose to describe their experiences. There are many instances where an author expresses an opinion without attaching any positive or negative affectively charged words (e.g., “the computer battery only last three hours”). These instances are more challenging to categorize as positive or negative as they require a large body of knowledge, but their categorization is not impos- sible. Taking the above statement as an example, “The computer battery only last three hours,” the word “only” is characterized by Sokolova and Lapalme [35] as a descriptive word. Depending on the context, the word “only” can serve as a hint for both a positive and a negative opinion. In our exam- ple, it indicates that the user of the computer expected the battery to last longer. Thus, the whole statement can be categorized as a negative opinion. Unlike lists of affectively charged words that work at word-level, the analysis of descriptive words requires a wider context, such as a whole sentence, a whole paragraph, or whole text. The analysis of descriptive words is also dependent on the topic of the text [35].

Thus, equipped with the knowledge of how opinions are differentiated from the other subjective terms, we for- mulate a schematic structure of an opinion that can guide future research in OM, as shown in Fig. 5. However, this structure does not describe opinions expressed as sar- casm or irony in which for instance, a seemingly positive opinion communicates the opposite of what is expressed (e.g., “It’s not that there isn’t anything positive to say about the field. There is. After 92 minutes, it ends” [72]). For efforts on detecting irony from text see Reyes and Rosso [72].

Using a simple scenario, we will illustrate how all these proposed aspects of an opinion can appear and be anno- tated in text.

6 DISCUSSION AND FUTURE WORK

This paper focused on highlighting the differences between five subjective terms and encouraging the NLP community to treat them differently. During our research, we faced challenges that could be further exploited as future work.

In compiling the definitions of the subjective terms, we noted that affect, emotions, feelings, and sentiments have theoretical underpinnings from evolutionary theory, psy- chology, folk psychology, and cognitive, neurological, phys- iological, and social constructivism (e.g., [53], [73], [74], [75], [76]). Given that all these theories may seem to contradict one another, a careful look shows that they just focus on dif- ferent aspects of the same concept. For example, evolution- ary theorists view affect, emotions, feelings, and sentiments as being shaped by evolution to fulfill survival-benefiting functions [77] whereas social constructivists view them and their meaning as a product of culture and learned social rules [75]. We presented a compilation of the definitions of the affective terms that are of interest from the computa- tional linguistics perspective. However, the convergence of the different psychological and cognitive theories upon which our definitions are based remains an interesting research venture.

Furthermore, for practical reasons, we restricted our work to the five subjective terms that are more frequently researched in NLP, but we recognized that there are other terms that also belong to this categorization, for example, attitude, mood, sensation, and temperament. As the fine- grained categorization of the terms increases, their distinc- tions also become fuzzier. Thus, as future work, using the differentiating factors uncovered in this work, we plan to investigate on a common framework for detecting other subjective terms.

An aspect that we did not investigate in this paper is what analytic philosophy calls the use-mention distinction [78]. That is, there is a difference between an author using a word and mentioning the word. “Use of an expression involves reference to what the expression refers to; mention of an expression involves reference to the expression itself” [78]. Put simply, use is the application of the word meaning (e.g., I like chocolate), whereas mention is a statement about

Fig. 5. Schematic structure on an opinion.

108 IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. 5, NO. 2, APRIL-JUNE 2014

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

the word per se (e.g., chocolate has two letters). This is a chal- lenge for NLP techniques applied to analyzing language usage similar to detecting sarcasm, irony, exaggerations, and understatements. The use-mention distinction is impor- tant, for instance, when a researcher is interested in flagging authors who are abusive. The researcher might collect the mention of the word “shit” as it is offensive, but the writer might mention it without using it to be offensive (see Wilson [79] on the use-mention distinction in NLP).

Research remains open to dive deeply into how the dif- ferent levels of analysis (word, phrase, sentence, paragraph, document, community, society, culture, etc.) might affect the detection abilities of the subjective terms. The structures we provided are to serve as a guide for researchers to judge on the correct capture of the subjective terms.

7 CONCLUSION

This paper shed light on the differentiating factors between subjective terms and identified their fundamen- tal concepts. We further gave suggestions on how these concepts could be incorporated to advance their detection in natural language. The differentiating factors between affect, emotions, feelings, sentiments, and opinion revealed the following:

� Affect is non-conscious and is difficult to conceptual- ize in language, revealing that what can be detected from text is the conscious expression of affect, which we found to be feelings and emotions.

� Feelings are conscious phenomena that have been labeled and they can be detected from text.

� Emotions are complex psychological phenomena that are near impossible to detect in totality from text. What we were able to detect is the written con- scious experience of five factors (appraisals, feelings, physiological reactions, expressive behavior, and readiness to act in a certain way), which constitute emotions. We also found that the use of any words to convey emotions is influenced by culture. We fur- ther recommend the inclusion of ethnographic stud- ies to answer questions such as “What role does culture play in the linguistic expression of emotions?” Answering that would enable NLP researchers to create more robust ED detection algorithms.

� Sentiments are enduring emotional dispositions that have developed over time about particular objects. Conclusions about sentiments in text have to be per- formed for a period. In turn, this will also help in improving decision making.

� Opinions are personal interpretations of information, which may or may not be associated with an emotion or sentiment.

It is the recommendation of this paper that investiga- tions on affect, emotion, feeling, sentiment, and opinion in text should proceed with greater consideration into the differences in terminologies and the role played by cultural, social, and personal differentiating factors to select the proper type of analysis needed for the respec- tive tasks.

ACKNOWLEDGMENTS

This work was supported by the “Detecting and Visualizing Emotions and their Changes in Text” project (No. 14166), funded by the Academy of Finland.

REFERENCES [1] J. Wiebe, “Identifying subjective characters in narrative,” in Proc.

13th Int. Conf. Comput. Linguistics, 1990, vol. 2, pp. 401–406. [2] B. Pang and L. Lee, Opinion Mining and Sentiment Analysis. New

York, NY, USA: Cambridge Univ. Press, 2008. [3] E. Hovy, "What are subjectivity, sentiment and affect?" presented

at the Workshop Sentiment Analysis Where AI Meets Psychology, Chiang Mai, Thailand, 2011.

[4] A. Balahur, J. M. Hermida, and A. Montoyo, “Building and exploiting emotinet, a knowledge base for emotion detection based on the appraisal theory model,” IEEE Trans. Affective Com- put., vol. 3, no. 1, pp. 88–101, Jan.–Mar. 2012.

[5] C. O. Alm, “The role of affect in the computational modeling of natural language,” Language Linguistics Compass, vol. 6, no. 7, pp. 416–430, 2012.

[6] Merriam-Webster Online. (2014). [Online]. Available: http:// www.merriam-webster.com/dictionary/

[7] K. R. Scherer, “Psychological models of emotion,” in The Neuro- psychology of Emotion, J. Borod, ed., Oxford, U.K.: Oxford Univ. Press, 2000, pp. 137–167.

[8] B. Massumi, Parables for the Virtual: Movement, Affect, Sensation. Durham, NC, USA: Duke Univ. Press, 2002.

[9] E. Shouse, “Feeling, emotion, affect,” Media Culture J., vol. 8, no. 6, p. 1, 2005.

[10] K. S. Fleckenstein, “Defining affect in relation to cognition: A response to Susan McLeod,” J. Adv. Composition, vol. 11, no. 2, pp. 447–453, 1991.

[11] C. Batson, L. Shaw, and K. Oleson, “Differentiating affect, mood, and emotion: Toward functionally based conceptual distinctions,” Rev. Personality Soc. Psychology, vol. 13, pp. 294–326, 1992.

[12] P. A. Thoits, “The sociology of emotions,” Annu. Rev. Sociology, vol. 15, pp. 317–342, 1989.

[13] R. Zajonc, “Feeling and thinking: Preferences need no inferences,” Amer. Psychologist, vol. 35, no. 2, pp. 151–175, 1980.

[14] J. Friedenberg and G. Silverman, Cognitive Science: An Introduction to the Study of Mind. 1st ed. Newbury Park, CA, USA: SAGE, 2005.

[15] A. Wierzbicka, Emotions Across Languages and Cultures: Diver- sity and Universals. Cambridge, U.K.: Cambridge Univ. Press, 1999.

[16] P. R. Kleinginna and A. Kleinginna, “A categorized list of emotion definitions, with suggestions for a consensual definition,” Motiva- tion Emotion, vol. 5, no. 4, pp. 345–379, 1981.

[17] R. Dolan, “Emotion, cognition and behavior,” Science, vol. 298, no. 5596, pp. 1191–1194, 2002.

[18] R. Cattell, “Sentiment or attitude? The core of a terminology prob- lem in personality research,” J. Personality, vol. 9, pp. 6–17, 2006.

[19] S. L. Gordon, “The sociology of sentiments and emotion,” in Social Psychology: Sociological Perspectives, M. Rosenberg and R. H. Turner, eds., New York, NY, USA: Basic Books, 1981 pp. 562–592.

[20] H. Murray and C. Morgan, “A clinical study of sentiments i,” Genetic Psychological Monograph, vol. 32, pp. 153–311, 1945.

[21] C. D. Broad, “Emotion and Sentiment,” J. Aesthetics Art Criticism, vol. 13, no. 2, pp. 203–214, 1971.

[22] K. Young, Social Psychology: An Analysis of Social Behavior. New York, NY, USA: Alfred A. Knopf, 1930.

[23] M. Fishbein and I. Ajzen, “Attitudes and opinions,” Annu. Rev. Psychol., vol. 23, pp. 487–544, 1972.

[24] S. Kim and E. Hovy, “Determining the sentiment of opinions,” in Proc. 20th Int. Conf. Comput.l Linguistics, PA, USA, 2004, pp. 1367– 1363.

[25] M. Sokolova and V. Bobicev, “Sentiments and opinions in health- related web messages,” in Proc. Recent Adv. Natural Language Pro- cess., 2011, pp. 132–139.

[26] B. Liu, “Sentiment analysis and subjectivity,” in Handbook of Natu- ral Language Processing, 2nd ed., N. Indurkhya and F. J. Damerau, eds. Boca Raton, FL, USA: CRC Press, 2010.

[27] B. Massumi, “Notes on the translation and acknowledgements,” in A Thousand Plateaus: Capitalism and Schizophrenia (illustrated ed.). Minneapolis, MN, USA: Univ. of Minnesota Press, 1987.

MUNEZERO ET AL.: ARE THEY DIFFERENT? AFFECT, FEELING, EMOTION, SENTIMENT, AND OPINION DETECTION IN TEXT 109

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

[28] I. Matthis, “Sketch for a metapsychology of affect,” Int. J. Psycho- anal., vol. 81, pp. 215–227, 2000.

[29] J. Kagan, “On emotion and its development: A working paper,” in The Development of Affect, M. Lewis and L. Rosenblum, eds., New York, NY, USA: Plenum, 1978 pp. 11–42.

[30] N. Besnier, “Language and Affect,” Annu. Rev. Anthropol., vol. 19, no. 1, pp. 419–451, 1990.

[31] J. E. Stets, “Emotions and sentiments,” in Handbook of Social Psy- chology, J. Delamater, ed., New York, NY, USA: Plenum, 2003, pp. 309–335,.

[32] R. J. Davidson, K. R. Scherer, and H. H. Goldsmith, in Handbook of Affective Science. 1st ed. New York, NY, USA: Oxford Univ. Press, 2002.

[33] A. Ben-Ze’ev, The Subtlety of Emotions, vol. 12, Cambridge, MA, USA: MIT Press, 2000.

[34] J. A. Russell and L. F. Barrett, “Core affect, prototypical emo- tional episodes, and other things called emotion: Dissecting the elephant,” J. Personality Soc. Psychol., vol. 76, no. 5, pp. 805–819, 1999.

[35] M. Sokolova and G. Lapalme, “Learning opinions in user-gener- ated web content,” Natural Language Learn., vol. 17, no. 4, pp. 541– 567, 2011.

[36] P. Lasersohn, “Context dependence, disagreement, and predicates of personal taste,” Linguistics Philosophy, vol. 28, pp. 643–686, 2005.

[37] R. W. Picard, Affective Computing. Cambridge, MA, USA: MIT Press, 1997.

[38] S. Schachter, “Cognition and centralist-peripheralist controversies in motivation and emotion,” in Handbook of Psychobiology, M. S. Gazzaniga and C. B. Blakemore, eds., New York, NY, USA: Aca- demic, 1975 pp. 529–564.

[39] P. N. Johnson-Laird and K. Oatley, “The language of emotions: An analysis of a semantic field,” Cognition Emotion, vol. 3, pp. 81–123, 1989.

[40] K. R. Scherer, “Emotion as a multicomponent process: A model and some cross-cultural data,” Rev. Personality Soc. Psychol., vol. 5, pp. 37–63, 1984.

[41] R. Reisenzein, “On oatley and Johnson-Laird’s theory of emotion and hierarchical structures in the affective lexicon,” Cognition Emotion, vol. 9, no. 4, pp. 383–416, 1995.

[42] N. Frijda, The Emotions. New York, NY, USA: Cambridge Univ. Press, 1986.

[43] K. R. Scherer, “Appraisal considered as a process of multilevel sequential processing,” in Appraisal Processes in Emotion: Theory, Methods, Research, K. R. Scherer, A. Schorr, and T. Johnstone, eds., New York, NY, USA: Oxford Univ. Press, 2001.

[44] R. A. Calvo and S. D’Mello, “Affect detection: An interdisciplinary review of models, methods, and their applications,” IEEE Trans. Affective Comput., vol. 1, no. 1, pp. 18–37, Jan. 2010.

[45] B. Mesquita and J. Leu, “The cultural psychology of emotion,” in Handbook of Cultural Psychology, S. Kitayama and D. Cohen, eds., New York, NY, USA: Guilford Press, 2007 pp. 734–759.

[46] C. Conati, “Probabilistic assessment of user’s emotions during the interaction with educational games,” Appl. Artif. Intell., vol. 16, no. 7/8, pp. 555–575, 2002.

[47] A. Ackerman and B. Puglisi, The Emotion Thesaurus: A Writer’s Guide to Character Expression. Los Angeles, CA, USA: CreateSpace Independent Publishing Platform, 2012.

[48] N. Frijda, “Emotions and emotion words,” in Everyday Conceptions of Emotion: An Introduction to the Psychology, Anthropology and Lin- guistics of Emotion, J. A. Russell, J. Fern�andez-Dolls, A. S. R. Man- stead, and J. C. Wellenkamp, eds., Norwell, MA, USA: Kluwer, 1995.

[49] J. Read, D. Hope, and J. Carroll, “Annotating expressions of appraisal in English,” in Proc. Linguistic Annotation Workshop, 2007, pp. 93–100.

[50] J. Reilly and L. Seibert, “Language and emotion,” in Handbook of Affective Sciences, Series in Affective Science, R. J. Davidson, K. R. Scherer, and H. H. Goldsmith, eds. New York, NY, USA: Oxford Univ. Press.

[51] M. Ostendorf, E. Shriberg, and A. Stolcke, “Human language tech- nology: Opportunities and challenges,” in Proc. IEEE Int. Conf. Acoustics, Speech, Signal Process., 2005, vol. 5, pp. 949–952.

[52] D. Derks, A. H. Fisher, and A. E. R. Bos, “The role of emotion in computer-mediated communication: A review,” Comput. Human Behavior, vol. 24, pp. 766–785, 2008.

[53] P. Ekman, “Are there basic emotions?” Psychol. Rev., vol. 99, no. 3, pp. 550–553, 1992.

[54] J. A. Russell, “Evidence for a three-factor theory of emotions,” J. Res. Personality, vol. 11, no. 3, pp. 273–294, 1977.

[55] A. Ortony, G. L. Clore, and A. Collins, The Cognitive Structure of Emotions. Cambridge U.K.: Cambridge Univ. Press, 1988.

[56] M. Sch€oder, “Speech and emotion research: An overview of research frameworks and a dimensional approach to emotional speech synthesis,” PhD dissertation, Inst. Phonetics, Saarland Univ., Saarbr€ucken, Germany, 2004.

[57] C. Strapparava and A. Valitutti, “WordNet-Affect: An affective extension of WordNet,” in Proc. 4th Int. Conf. Language Resources Eval., 2004, pp. 1083–1086.

[58] M. M. Bradley and P. J. Lang, “Affective norms for English words (ANEW): Instruction manual and affective ratings,” The Center for Research in Psychophysiological, Univ. Florida, Gainesville, FL, USA, Tech. Rep. C-1, 1999.

[59] A. Balahur, J. M. Hermida, and H. Tanev, “Detecting implicit emotion expressions from text using ontological resources and lexical learning,” in New Trends of Research in Ontologies and Lexical Resources, A. Oltramari, P. Vossen, L. Qin, and E. Hovy, eds., New York, NY, USA: Springer, 2013, pp. 235–255,.

[60] A. Das, “PsychoSentiWordNet,” in Proc. ACL Student Session, 2011, pp. 52–57.

[61] Facebook. (2014). [Online]. Available: https://www.facebook. com.

[62] V. V. French, “The structure of sentiments: I. A restatement of the theory of sentiments,” J. Personality, vol. 15, no. 4, pp. 247–282, 1947.

[63] R. K. Shelly, “Emotions, sentiments and performance expect- ations,” in Theory and Research on Human Emotions (Advances in Group Processes, vol. 21, J. H. Turner, ed. Bradford, U.K.: Emerald Group Publishing, 2004, pp. 141–165..

[64] R. Plutchik, “A general psychoevolutionary theory of emotion,” in Emotion: Theory, Research, and Experience: Vol. 1 Theories of Emotion, R. Plutchik and H. Kellerman, eds., New York, NY, USA: Aca- demic, 1980, pp. 3–33,.

[65] M. Munezero, M. Mozgovoy, C. Suero Montero and E. Sutinen, “Identifying enduring sentiments with emotwitter—A fine- grained visualization system of sentiments in twitter posts,” unpublished.

[66] A. Sayeed, “An opinion about opinions about opinions: Subjectiv- ity and the aggregate reader,” in Proc. North Amer. Chapter Assoc. Comput. Linguistics: Human Language Technol., Atlanta, Georgia, 2013, pp. 691–696.

[67] G. A. Miller, “WordNet: A lexical database for English,” Commun. ACM, vol. 38, no. 11, pp. 39–41, 1995.

[68] P. D. Turney, “Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews,” in Proc. 40th Annu. Meeting Assoc. Comput. Linguistics, Stroudsburg, PA, USA, 2002, pp. 417–424.

[69] J. Kamps, M. Marx, R. Mokken, and M. de Rijke, “Using WordNet to measure semantic orientation of adjectives,” in Proc. 4th Int. Conf. Language Resources Eval., 2004, pp. 1115–1118.

[70] E S. Kim and E. Hovy, “Automatic detection of opinion bearing words and sentences,” in Proc. 2nd Joint Int. Conf. Natural Language Process., 2005, pp. 61–66.

[71] E. Riloff and J. Wiebe, “Learning extraction patterns for subjective experiences,” in Proc. Empirical Methods Natural Language Process., 2003, pp. 105–112.

[72] A. Reyes and P. Rosso, “Making objective decisions from subjec- tive data: Detecting irony in customer reviews,” Decision Support Syst., vol. 53, no. 4, pp. 754–760, 2012.

[73] W. James, “What is an emotion?” Mind, vol. 9, no. 34, pp. 188–205, 1884.

[74] R. S. Lazarus, Emotion and Adaptation. New York, NY, USA: Oxford Univ. Press, 1991.

[75] J. R. Averill, “A constructivist view of emotion,” in Emotion: The- ory, Research and Experience, vol. 1, R. Plutchik and H. Kellerman, Eds. New York, NY, USA: Academic, 1980, pp. 305–339.

[76] K. Oatley, Best Laid Schemes: The Psychology of the Emotions. Cam- bridge, U.K.: Cambridge Univ. Press, 1992.

[77] P. Ekman and W. V. Friesen, Unmasking the Face: A Guide to Recog- nizing Emotions from Facial Expressions. Englewood Cliffs, NJ, USA: Prentice-Hall, 1975.

[78] D. Sperber and D. Wilson, “Irony and the use-mention dis- tinction,” in Radical Pragmatics, P. Cole, ed., New York, NY, USA: Academic, 1981, pp. 295–318,.

110 IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. 5, NO. 2, APRIL-JUNE 2014

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.

[79] S. Wilson, “A computation theory of the use-mention distinction in natural language,” PhD dissertation, Faculty of the Graduate School of the Univ. Maryland, College Park, MD, USA, 2011.

Myriam D. Munezero received the BS degree in computer science and mathematics from the University of Namibia, and the MS degree in computer science from the University of Joensuu. She is currently working toward the PhD degree at the University of Eastern Finland (UEF). She has worked for four years as a junior researcher at UEF in a project funded by the Academy of Finland. She is cur- rently a coordinator at UEF. Her research interests include the area of natural language

processing, particularly on detecting emotions from text. Her research in this area has resulted in four conference papers, one co-authored conference paper and two journal papers.

Calkin Suero Montero received the engineering master’s degree and computer science PhD degree from Hokkaido University, Japan. She is currently a post-doctoral research fellow at the University of Eastern Finland. She has expertise in human computer interaction and also has extensive experience working within international environments in the fields of computational lin- guistic, sentiment analysis, novel interaction techniques and usability and social acceptance of technology. She has received two Best Paper

Awards from the IEEE Sapporo Section and her research has been pub- lished in world level international forums including MobileHCI and Coling/ACL. She is a member of the IEEE, the IEEE Computer Society and IEEE WIE Society.

Erkki Sutinen received the PhD degree in com- puter science, and the MSc degree in mathemat- ics from the University of Helsinki. He is currently the head of the education technology research group at the University of Eastern Finland. His research interests include technologies for spe- cial education, contextualized IT education, visual and textual tools for learning, and natural language processing methods for business intel- ligence and for emotion detection. He has pub- lished and supervised many students in the

above fields. He has received several research grants from the Acad- emy of Finland, The Finnish Funding Agency for Technology and Inno- vation (TEKES) and the European Union.

John Pajunen received the master’s degree in philosophy in 2008. He is currently employed at the University of Jyv€askyl€a as a university teacher. His current research interests include argumentation theory, logic, and philosophy of science. He is a member of IACAP.

" For more information on this or any other computing topic, please visit our Digital Library at www.computer.org/publications/dlib.

MUNEZERO ET AL.: ARE THEY DIFFERENT? AFFECT, FEELING, EMOTION, SENTIMENT, AND OPINION DETECTION IN TEXT 111

Authorized licensed use limited to: University of the Cumberlands. Downloaded on September 25,2021 at 02:29:30 UTC from IEEE Xplore. Restrictions apply.