About this Research Topic
Emotions, central to human cognition, play a major role in the way feelings and opinions are formed. They impact perception, memorization, learning, and so on. If they may be transmitted by specific lexical units, emotions are also based on fuzzy elements whose perception may differ from one person to another, depending on the context and socio-cultural aspects. Emotions are the subject of research in neuroscience, psychology, and sociology. These fields can contribute to NLP, for example in the analysis of social network text, and NLP can conversely contribute to the study of behaviors and the impact of emotions on children in learning situations, the study of sensory or perceptual disorders in mental health, and cognitive therapies targeting emotional recognition and social cognition. Lastly, emotion processing can transform the uses of IT systems which will have a strong societal impact, therefore it is imperative that it is fair and ethical.
The problems associated with the processing of emotions depend on whether the approach is for text analysis or text generation. In the first case, emotion processing consists of the ability to detect the textual areas carrying emotions and to extract the dominant emotion(s) or the emotional valence and intensity. Emotions can be carried by very factual information or by a strictly emotional vocabulary. This diversity of forms favors machine learning approaches that can combine very diverse features from representations of emotions in categories (multilabel classification problem) or according to several dimensions (regression problem). In the case of machine translation, the objective is to ensure that the texts in both languages have the same emotional signature. The objective of this Research Topic is to present the most recent and efficient solutions in terms of their ability to take into account the subjective character of emotions, adapt to contexts, and generalize to different languages and thematic domains.
The scope of this Research Topic is to address emotion detection and its integration on NLP tasks. Contributions should address the inter-individual variability in the perception of emotions and the capacity of current neural architectures and pre-trained models to generalize well. In addition, themes of privacy and diversity in datasets and their annotations should be explored. Disclosing an individual's emotions to help build models is exposing a part of their privacy and, without reliable anonymization, could lead to the creation of psychological profiles used against people, especially when annotations are accompanied by physical or physiological data.
Furthermore, an area of interest is the limits of considering text alone vs multimodal and physiological data. Since emotions can evolve quickly, the temporal dimension should be integrated to lead to relevant emotional signatures and since several emotions can be felt simultaneously, classifications should be multilabel. Another question concerns the interrelation between emotions, feelings, moods, and opinions.
Articles should present new approaches or employ traditional approaches to new tasks or in novel contexts. Articles that highlight the impact of interdisciplinary work or promising perspectives are encouraged.
Topics of interest include but are not limited to:
• Emotion processing in generating texts
• Emotion processing of multimodal and physiological data
• Analysis of messages on social networks, including emojis
• Searching for emotionally related content
• Generalization capacity of models
• Ethics and fairness of emotion-based NLP systems (privacy, security, transparency, robustness, diversity)
• Neural networks and hybrid approaches
• Emotion oriented recommenders and IR systems
• Identification of threatening messages or depression levels
The problems associated with the processing of emotions depend on whether the approach is for text analysis or text generation. In the first case, emotion processing consists of the ability to detect the textual areas carrying emotions and to extract the dominant emotion(s) or the emotional valence and intensity. Emotions can be carried by very factual information or by a strictly emotional vocabulary. This diversity of forms favors machine learning approaches that can combine very diverse features from representations of emotions in categories (multilabel classification problem) or according to several dimensions (regression problem). In the case of machine translation, the objective is to ensure that the texts in both languages have the same emotional signature. The objective of this Research Topic is to present the most recent and efficient solutions in terms of their ability to take into account the subjective character of emotions, adapt to contexts, and generalize to different languages and thematic domains.
The scope of this Research Topic is to address emotion detection and its integration on NLP tasks. Contributions should address the inter-individual variability in the perception of emotions and the capacity of current neural architectures and pre-trained models to generalize well. In addition, themes of privacy and diversity in datasets and their annotations should be explored. Disclosing an individual's emotions to help build models is exposing a part of their privacy and, without reliable anonymization, could lead to the creation of psychological profiles used against people, especially when annotations are accompanied by physical or physiological data.
Furthermore, an area of interest is the limits of considering text alone vs multimodal and physiological data. Since emotions can evolve quickly, the temporal dimension should be integrated to lead to relevant emotional signatures and since several emotions can be felt simultaneously, classifications should be multilabel. Another question concerns the interrelation between emotions, feelings, moods, and opinions.
Articles should present new approaches or employ traditional approaches to new tasks or in novel contexts. Articles that highlight the impact of interdisciplinary work or promising perspectives are encouraged.
Topics of interest include but are not limited to:
• Emotion processing in generating texts
• Emotion processing of multimodal and physiological data
• Analysis of messages on social networks, including emojis
• Searching for emotionally related content
• Generalization capacity of models
• Ethics and fairness of emotion-based NLP systems (privacy, security, transparency, robustness, diversity)
• Neural networks and hybrid approaches
• Emotion oriented recommenders and IR systems
• Identification of threatening messages or depression levels
Keywords: Neural networks and hybrid approaches, Emotion classification and models, Emotion extraction and detection, datasets and linguistic resources, Emotion oriented recommenders and IR systems, subjectivity, Emojis, Social networks, aided communication, text generation, identification of threatening messages or depression levels, security, transparency, robustness, Ethics and Fairness of Emotion based NLP systems, privacy, accepting diversity
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.
