Emotion Detection Task Research Articles

MultiEmo: Multi-task framework for emoji prediction

Emojis are becoming a new visual and linguistic tool that allows users to express their feelings and communicate with each other on social media. Driven by the importance of emoji interpretation, an emoji prediction task, which aims to predict the most likely emojis within a text, has gained significant attention. Thus, we propose a new deep neural network model for this task, MultiEmo (multi-task framework for emoji prediction), to predict the most relevant emoji by considering the emotion detection task. Our experiment shows that MultiEmo is superior to existing models using the Twitter dataset, implying that the models can learn richer representations from semantically related tasks. We systematically confirm that each emoji is associated with a particular emotion in a similar context. We also introduce new evaluation metrics to measure the comprehensive performance of the models.

How therapists' emotion recognition relates to therapy process and outcome.

Empathy is an essential characteristic for therapists that explained 9% of the variance in treatment outcomes. Many measures of empathy are based on self-report questionnaires. Therefore, they reflect how a person perceives his empathic abilities, which can be biased from his true abilities. The ability to recognize emotions has been found to be related to empathy and serves as a measure for the use of empathic abilities. In this study, we examined therapists' empathic abilities, using a novel task for recognizing emotions, and looked at how they related to the therapy process and outcomes. The study included 33 patient-therapist dyads. Therapist empathy was assessed with an emotion detection task, the JeFEE. Clients filled questionnaires after each session regarding therapy progress and their symptoms. We found that emotion recognition moderated the change in (a) client secure attachment to therapist, (b) client avoidant attachment to therapist, (c) client working alliance, (d) client rate of tense or upset they felt during the session, (e) client lack of emotional clarity of emotions, (f) client non-acceptance of emotional responses, (g) client overall emotion regulation and (h) client main target complaint. Implication for therapy and therapists' selection and training are discussed.

Relationships between social withdrawal and facial emotion recognition in neuropsychiatric disorders

BackgroundEmotion recognition constitutes a pivotal process of social cognition. It involves decoding social cues (e.g., facial expressions) to maximise social adjustment. Current theoretical models posit the relationship between social withdrawal factors (social disengagement, lack of social interactions and loneliness) and emotion decoding. ObjectiveTo investigate the role of social withdrawal in patients with schizophrenia (SZ) or probable Alzheimer's disease (AD), neuropsychiatric conditions associated with social dysfunction. MethodsA sample of 156 participants was recruited: schizophrenia patients (SZ; n = 53), Alzheimer's disease patients (AD; n = 46), and two age-matched control groups (SZc, n = 29; ADc, n = 28). All participants provided self-report measures of loneliness and social functioning, and completed a facial emotion detection task. ResultsNeuropsychiatric patients (both groups) showed poorer performance in detecting both positive and negative emotions compared with their healthy counterparts (p < .01). Social withdrawal was associated with higher accuracy in negative emotion detection, across all groups. Additionally, neuropsychiatric patients with higher social withdrawal showed lower positive emotion misclassification. ConclusionsOur findings help to detail the similarities and differences in social function and facial emotion recognition in two disorders rarely studied in parallel, AD and SZ. Transdiagnostic patterns in these results suggest that social withdrawal is associated with heightened sensitivity to negative emotion expressions, potentially reflecting hypervigilance to social threat. Across the neuropsychiatric groups specifically, this hypervigilance associated with social withdrawal extended to positive emotion expressions, an emotional-cognitive bias that may impact social functioning in people with severe mental illness.

Children's emotion perception in context: The role of caregiver touch and relationship quality.

From early in development, attention to emotional facial expressions is biased toward threat. We examined whether caregiver availability affects children's emotion perception and possibly reduces threat bias. In 3 experiments, 4- and 5-year-old children were asked to detect threatening (angry) and nonthreatening (happy) facial expressions on a touch screen. Children completed the emotion detection task either with or without their caregiver sitting next to them on a bench, first using a between-subjects design where each child completed the task once (Experiment 1; N = 40) and then using a within-subjects design where each child completed the task twice (Experiment 2; N = 20). In Experiments 1 and 2, mere caregiver presence had no effect on children's emotion detection or threat bias. In Experiment 3, we increased the salience of caregiver presence by having children complete the task while holding hands with their caregiver (Experiment 3; N = 45). In dyads reporting high quality child-caregiver relationships, handholding facilitated detection of happy faces and significantly reduced children's threat bias. These findings suggest that caregiver touch impacts emotion perception in children as a function of relationship quality and that caregiver context plays a role in shaping children's responses to their social environment. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

The Impact of Translating Resource-Rich Datasets to Low-Resource Languages Through Multi-Lingual Text Processing

Urdu is still considered a low-resource language despite being ranked as world’s $10^{th}$ most spoken language with nearly 230 million speakers. The scarcity of benchmark datasets in low-resource languages has led researchers to utilize more ingenious techniques to curb the issue. One such option widely adopted is to use language translation services to replicate existing datasets from resource-rich languages such as English to low-resource languages, such as Urdu. For most natural language processing tasks, including polarity assessment, words translated via Google translator from one language to another often change the meaning. It results in a polarity shift causing the system’s performance degradation, particularly for sentiment classification and emotion detection tasks. This study evaluates the effect of translation on the sentiment classification task from a resource-rich language to a low-resource language. It identifies and enlists words causing polarity shift into five distinct categories. It further finds the correlation between the language with similar roots. Our study shows 2-3 percentage points performance degradation in sentiment classification due to polarity shift as a result of translation from resource-rich languages to low-resource languages.

Affect detection from arabic tweets using ensemble and deep learning techniques

Affect detection from text has captured the attention of researchers recently. This is due to the rapid use of social media sites (e.g. Twitter, Facebook), which allows users to express their feelings, emotions, and thoughts in textual format. Analyzing emotion-rich textual data of social networks has many real-life applications. The context of an emotional text can be measured by analyzing certain features of this rich source of emotional information. Classifying text into emotional labels/intensities is considered a difficult problem. This paper resolves one of the state-of-the-art NLP research emotion and intensity detection tasks using Deep Learning and ensemble implementations. In this paper, we developed several innovative approaches; (a) bidirectional GRU_CNN (BiGRU_CNN), (b) conventional neural networks (CNN), and (c) XGBoost regressor (XGB). The ensemble of BiGRU_CNN, CNN, and XGB is used to solve an emotion intensity (EI-reg) task of the SemEval-2018 Task1 (Affect in Tweets). Our proposed ensemble approach was evaluated using a reference dataset of the SemEval-2018 Task1. Results show that our approach is well above the baseline for this task. It also achieved a Pearson of (69.2%), with an enhancement of 0.7% in comparison with previous best performing models.

Topic-Enhanced Capsule Network for Multi-Label Emotion Classification

Identifying multiple emotions in a piece of text is an important research topic in the NLP community.Existing methods usually model the task as a multi-label classification problem, while these work has two issues. First, these methods fail to leverage the topic information of the text, which has been shown to be effective for sentiment analysis task. Second, different parts of the text can contribute differently to predicting different emotion labels, so the proposed model needs to capture effective features for each corresponding emotion, which is not considered by existing models. To tackle these problems, we propose a topic-enhanced capsule network, which contains two main parts: a variational autoencoder and a capsule module, for multi-label emotion detection task. Specifically, the variational autoencoder can learn the latent topic information of the text, and the capsule module can capture rich features for corresponding emotion. Experimental results on two benchmark datasets show that the proposed model achieves the current best performance, outperforming previous methods and strong baselines by a large margin.

Kāvi: An Annotated Corpus of Punjabi Poetry with Emotion Detection Based on ‘Navrasa’

As the poetries are penned with numerous emotions and feelings, the computational linguistic analysis for emotion detection from poetry is a very challenging task. The main aim of this work is to detect emotions from Punjabi poetries based on different features present in Punjabi poetries. This is a pioneering approach in emotion detection from Punjabi poetries. ‘Kāvi’ Punjabi poetry corpus was manually annotated based on the Indian concept of ‘Navrasa’. In this corpus, 948 poetries were classified into 9 emotion states as presented in ‘Navrasa’, namely ‘karuna’, ‘shringar’, ‘hasya’, ‘raudra’, ‘veer’, ‘bhayanak’, ‘vibhata’, ‘adbhut’, and ‘shaanti’. This corpus was manually annotated and the Kappa Fleiss index was calculated for inter-annotator agreement. The different features (linguistic, poetic and statistical) were used for building the classifier and the two machine learning techniques, Naïve Bayes (NB) and Support Vector Machine (SVM) were experimented with these features. It was found that with an accuracy of 70.02%, SVM improved the overall accuracy of the emotion classification task. Further, Poetic features were found to outperform the linguistic features for the emotion detection task.

An effective approach for emotion detection in multimedia text data using sequence based convolutional neural network

In the recent trends, the world has stepped into a multimedia era for enhancing business, recommendation systems, and information retrieval, etc. Multimedia data is highly rich in contents which express different human emotions. Several issues for emotion detection from multimedia images & videos have been addressed in this domain, but a very less effort has been applied for text data. The evaluation of deep learning has outperformed traditional techniques in sentiment analysis tasks. Inspired by the work done in the field of sentiment analysis, a deep learning based framework has been implemented on multimedia text data for the task of fine-grained emotion detection. The presented work introduces a new corpus which expresses different forms of emotions collected from a TV show’s transcript. A manual annotation of the corpus has been conducted with the help of English expert annotators. As an emotion detection framework, this paper proposes a sequence-based convolutional neural network(CNN) with word embedding to detect the emotions. An attention mechanism is applied in the proposed model which allows CNN to focus on the words that have more effect on the classification or the part of the features that should be attended more. The main aim of the work is to develop a framework such a way to generalize to newly collected data and help business to understand the customer’s mind and social media monitoring as it allows us to gain an overview of the wider public opinion behind certain topics. Experiments conducted on the dataset shows that the proposed framework correctly detects the emotions from the text with good precision and accuracy score.

ShEMO: a large-scale validated database for Persian speech emotion detection

This paper introduces a large-scale, validated database for Persian called Sharif Emotional Speech Database (ShEMO). The database includes 3000 semi-natural utterances, equivalent to 3 h and 25 min of speech data extracted from online radio plays. The ShEMO covers speech samples of 87 native-Persian speakers for five basic emotions including anger, fear, happiness, sadness and surprise, as well as neutral state. Twelve annotators label the underlying emotional state of utterances and majority voting is used to decide on the final labels. According to the kappa measure, the inter-annotator agreement is 64% which is interpreted as “substantial agreement”. We also present benchmark results based on common classification methods in speech emotion detection task. According to the experiments, support vector machine achieves the best results for both gender-independent (58.2%) and gender-dependent models (female = 59.4%, male = 57.6%). The ShEMO will be available for academic purposes free of charge to provide a baseline for further research on Persian emotional speech.

Emotion Detection Deficits and Decreased Empathy in Patients with Alzheimer's Disease and Parkinson's Disease Affect Caregiver Mood and Burden.

Background: Changes in social cognition occur in patients with Alzheimer’s disease (AD) and Parkinson’s disease (PD) and can be caused by several factors, including emotion recognition deficits and neuropsychiatric symptoms (NPS). The aims of this study were to investigate: (1) group differences on emotion detection between patients diagnosed with AD or PD and their respective caregivers; (2) the association of emotion detection with empathetic ability and NPS in individuals with AD or PD; (3) caregivers’ depression and perceived burden in relation to patients’ ability to detect emotions, empathize with others, presence of NPS; and (4) caregiver’s awareness of emotion detection deficits in patients with AD or Parkinson.Methods: In this study, patients with probable AD (N = 25) or PD (N = 17), and their caregivers (N = 42), performed an emotion detection task (The Awareness of Social Inference Test—Emotion Evaluation Test, TASIT-EET). Patients underwent cognitive assessment, using the Behavioral Neurology Assessment (BNA). In addition, caregivers completed questionnaires to measure empathy (Interpersonal Reactivity Index, IRI) and NPS (Neuropsychiatric Inventory, NPI) in patients and self-reported on depression (Geriatric Depression Scale, GDS) and burden (Zarit Burden Interview, ZBI). Caregivers were also interviewed to measure dementia severity (Clinical Dementia Rating (CDR) Scale) in patients.Results: The results suggest that individuals with AD and PD are significantly worse at recognizing emotions than their caregivers. Moreover, caregivers failed to recognize patients’ emotion recognition deficits and this was associated with increased caregiver burden and depression. Patients’ emotion recognition deficits, decreased empathy and NPS were also related to caregiver burden and depression.Conclusions: Changes in emotion detection and empathy in individuals with AD and PD has implications for caregiver burden and depression and may be amenable to interventions with both patients and caregivers.

Sentence-level emotion mining based on combination of adaptive Meta-level features and sentence syntactic features

Nowadays, because of increasing of text data, recognizing the emotions of text can help to get a better comprehension of context. However, finding emotional information from text is a very complex task because it is needed to automatically understand of human sentences that usually are vague and dependent on context which should be interpreted and represented in different ways. The sense of a word can be inferred by investigating the frequency of occurrence of the word in a large corpus of annotated text. This paper has presented a method for learning adaptive lexicon from existing lexicon resources (static lexicons) to improve performance of emotion detection task for two data sets. The learning of adaptive lexicon would allow us to distinguish between the initial emotion of words in the static lexicons and adaptive emotion of words that is mentioned in context, and we get a better understanding of the emotional orientation of words. Furthermore, this study proposes a novel approach for emotion detection based on the combination of Meta-level features derived from static and adaptive lexicons and sentences syntactic features. To the best of our knowledge, this is the first study that provides a comprehensive analysis of the relative importance of a very diverse feature set for automatic emotion detection. Extensive experiments on ISEAR and Aman data sets show that learning adaptive lexicon enables emotion mining algorithms to be more accurate.

Effects of Magnification on Emotion Perception in Patients With Age-Related Macular Degeneration.

Individuals with low vision often experience difficulties in performing tasks of daily living, such as face perception. This leads them to having difficulties with social interactions, as they can no longer correctly perceive the emotion of others. The present study investigated the effects of magnification on face perception in participants with age-related macular degeneration (AMD), and their ability to detect and categorize emotions. It was hypothesized that patients with AMD would be less accurate in comparison to healthy controls, but that magnification would improve their performance to that of controls. Faces containing happy, angry, or neutral emotion were both doubled (equivalent of arm's length distance) and decreased by half in size (equivalent of across the street). The ability to detect and to discriminate emotional content was compared between 20 AMD patients and 7 age-matched controls. Eye movements were recorded while conducting both tasks. Regardless of stimulus size, when compared to controls, we observed that individuals with AMD consistently performed with lower accuracy in both emotion detection and categorization tasks. Moreover, having images undergo a 2-fold increase in size did improve performance, but did not equate AMD participants' performance to that of the controls in either the emotion detection or categorization task. Eye movements in AMD participants were highly variable in position compared to controls. The data suggest that magnification alone does not appear to be the answer for improving emotion perception within individuals with low vision. Next steps should include an evaluation of the effects of viewing strategy.

Exploitation of phased-based features for emotional arousal evaluation from speech

The mel cepstral coefficients representing magnitude spectrum and Teager energy operators are often used as features in emotion recognition. The phase spectrum information is generally ignored. In this work an approach is proposed based on the use of group delay function from all pole models (APGD) to represent the phase information for the emotional arousal recognition from speech. The experiments were done on the CRISIS acted speech database with four levels of stress. The results of the arousal recognition system using the APGD features are compared to those using mel-frequency cepstral coefficients (MFCCs) and with Critical Band Based TEO Autocorrelation Envelope features (TEO-CB-Auto-Env) which have been successfully used in the task of emotion and stress detection in the past. The feature extraction is applied on the voiced parts of speech. The combination of APGD, MFCC, and TEO-CB-Auto-Env features has shown the best recognition results confirming the hypothesis that the phase and magnitude spectra contain complementary information and their combination can improve the reliability of the arousal recognition system.

Lexicon Generation for Emotion Detection from Text

General-purpose emotion lexicons (GPELs) that associate words with emotion categories remain a valuable resource for emotion detection. However, the static and formal nature of their vocabularies make them an inadequate resource for detecting emotions in domains that are inherently dynamic in nature. This calls for lexicons that are not only adaptive to the lexical variations in a domain but which also provide finer-grained quantitative estimates to accurately capture word-emotion associations. In this article, the authors demonstrate how to harness labeled emotion text (such as blogs and news headlines) and weakly labeled emotion text (such as tweets) to learn a word-emotion association lexicon by jointly modeling emotionality and neutrality of words using a generative unigram mixture model (UMM). Empirical evaluation confirms that UMM generated emotion language models (topics) have significantly lower perplexity compared to those from state-of-the-art generative models like supervised Latent Dirichlet Allocation (sLDA). Further emotion detection tasks involving word-emotion classification and document-emotion ranking confirm that the UMM lexicon significantly out performs GPELs and also state-of-the-art domain specific lexicons.

Emotion Discrimination Using Spatially Compact Regions of Interest Extracted from Imaging EEG Activity.

Lately, research on computational models of emotion had been getting much attention due to their potential for understanding the mechanisms of emotions and their promising broad range of applications that potentially bridge the gap between human and machine interactions. We propose a new method for emotion classification that relies on features extracted from those active brain areas that are most likely related to emotions. To this end, we carry out the selection of spatially compact regions of interest that are computed using the brain neural activity reconstructed from Electroencephalography data. Throughout this study, we consider three representative feature extraction methods widely applied to emotion detection tasks, including Power spectral density, Wavelet, and Hjorth parameters. Further feature selection is carried out using principal component analysis. For validation purpose, these features are used to feed a support vector machine classifier that is trained under the leave-one-out cross-validation strategy. Obtained results on real affective data show that incorporation of the proposed training method in combination with the enhanced spatial resolution provided by the source estimation allows improving the performed accuracy of discrimination in most of the considered emotions, namely: dominance, valence, and liking.

Perception of emotion in facial stimuli: The interaction of ADRA2A and COMT genotypes, and sex.

Emotional facial stimuli are important social signals that are essential to be perceived and recognized in order to make appropriate decisions and responses in everyday communication. The ability to voluntarily guide attention to perceive and recognize emotions, and react to them varies largely across individuals, and has a strong genetic component (Friedman et al., 2008). Two key genetic variants of the catecholamine system that have been related to emotion perception and attention are the catechol-O-methyl transferase genetic variant (COMT Val158Met) and the α2A-receptor gene promoter polymorphism (ADRA2A C-1291G) accordingly. So far, the interaction of the two with sex in emotion perception has not been studied. Multilevel modeling method was applied to study how COMT Val158Met, ADRA2A C-1291G and sex are associated with measures of emotion perception in a large sample of young adults. Participants (n=506) completed emotion recognition and behavioral emotion detection tasks. It was found that COMT Val158Met genotype in combination with the ADRA2A C-1291G and sex predicts emotion detection, and perception of valence and arousal. In simple visual detection, the ADRA2A C-1291G G-allele leads to slower detection of a highly arousing face (scheming), which is modulated by each additional COMT Val158Met Met-allele and male sex predicting faster responses. The combination of G-allele, Met-allele and male sex also predicts higher perceived negativity in sad faces. No effects of C-1291G, Val158Met, and sex were found on verbal emotion recognition. Applying the findings to study the interplay between catecholamine-O-methyl transferase activity and α2A-receptors in emotion perception disorders (such as ADHD, autism and schizophrenia) in men and women would be the next step towards understanding individual differences in emotion perception.

Using minimal number of electrodes for emotion detection using brain signals produced from a new elicitation technique

Emotion is an important aspect in the interaction between humans. There is a great interest for detecting emotions automatically. Current approaches for emotion detection using EEG are not practical for real-life situations because researchers ask participants to reduce any motion and facial muscle movement, reject noisy EEG data and rely on large number of electrodes. In this paper, we propose an approach that analyses highly contaminated brain signals. We then extract relevant features for the emotion detection task based on neuroscience findings. We reached an average accuracy of 51%, 53%, 58% and 61% for joy, anger, fear and sadness, respectively. We are also applying our approach on fewer number of electrodes that ranges from 4 to 25 electrodes and we reached an average classification accuracy of 33% for joy emotion, 38% for anger, 33% for fear and 37.5% for sadness using 4 or 6 electrodes only.

The Scorn of Status: A Bias Toward Perceiving Anger on High-Status Faces

The expression of anger is an important signal of both danger and threat to social relationships. Furthermore, anger expressed by high-status people should be especially apparent due to the fact that high-status people may be perceived as more able to act on their intentions than low-status people. In the current research, we test the hypothesis that a perceiver will be biased toward anger appearing on the face of a high-status compared to a low-status face. Using an emotion detection task across two experiments, our findings revealed that White participants perceived anger to persist longer (Study 1) and appear sooner (Study 2) on the faces of high-status compared to low-status targets. This suggests that social context, in general, and social status (as indicated by occupation), in particular, can influence the perception of anger.

Word and Sentence Level Emotion Analyzation in Telugu Blog and News

Emotion analysis, a recent sub discipline at the crossroads of information retrieval and computational linguistics is becoming increasinglyimportant from application view points of affective computing.Emotion is crucial to identify as it is not open to any objective observation or verification. In this paper, emotion analysis on blog texts has been carried out for a less privileged language, Teluguand the same system has been applied on the English SemEval 20 07 affect sensing corpus containing only news headlines. A set of six emotion tags, namely, happy ( ), sad ( ), anger ( ), fear ( ), surprise ( )and disgust ( ), have been selected towards this emotion detection task for re liable and semi-automatic annotation of blog and news data. Conditional Random Field (CRF) based classifier has been applied for recognizing six basic emotion tags for different words of a sentence. The classifier accuracy has been improved by arranging anequal distribution of emotional tags and non -emotional tag. A score based technique has been adopted to calculate and assign tag weights to each of the six emotion tags. A sense based scoring strategy has been applied to identify sentence level emotion sc ores for the six emotion tags based on the acquired word level emotion tags. Sentence level emotion tagging has been carried out based on the maximum obtained sentence level emotion scores. Evaluation has been conducted for each emotion class separately on200 test sentences from each of the Telugublog and English news data. The system has resulted accuracies of 69.82% and 71.06 % for happy, 70.24% and 66.42% for sad, 65.73% and 64.27% for anger, 76.01% and 69.90% for disgust, 72.19% and 73.59% for fear and 70.54% and 66.64% for surprise emotion classes on blog and news test data respectively.