TERMS: textual emotion recognition in multidimensional space

Abstract

Microblogs generate a vast amount of data in which users express their emotions regarding almost all aspects of everyday life. Capturing affective content from these context-dependent and subjective texts is a challenging task. We propose an intelligent probabilistic model for textual emotion recognition in multidimensional space (TERMS) that captures the subjective emotional boundaries and contextual information embedded in a text for robust emotion recognition. It is implausible with discrete label assignment;therefore, the model employs a soft assignment by mapping varying emotional perceptions in a multidimensional space and generates them as distributions via the Gaussian mixture model (GMM). To strengthen emotion distributions, TERMS integrates a probabilistic emotion classifier that captures the contextual and linguistic information from texts. The integration of these aspects, the context-aware emotion classifier and the learned GMM parameters provide a complete coverage for accurate emotion recognition. The large-scale experimentation shows that compared to baseline and state-of-the-art models, TERMS achieved better performance in terms of distinguishability, prediction, and classification performance. In addition, TERMS provide insights on emotion classes, the annotation patterns, and the models application in different scenarios.

Deep learning-based dimensional emotion recognition for conversational agent-based cognitive behavioral therapy

“…The performance gap between datasets with known and unknown heuristics underlines the importance of training prospective DL models on multiple datasets to promote a generalized understanding of emotions. Our model, furthermore, showed an improvement in Pearson correlation coefficient values for valence and arousal in comparison to the results of Ghafoor et al (2023) and respectively to the models that Ghafoor and colleagues compared their model to. Their performance measurements were however performed on a self-created, unpublished dataset and are therefore hard to reproduce.…”

“…When comparing the Pearson correlation coefficient to the state-of-the-art DL model for dimensional text-based emotion recognition by Park et al (2021) , the present model outperforms the one by Park et al in all VAD dimensions (by r = 0.06 for valence, r = 0.2 for arousal, and r = 0.12 for dominance, see Table 5 ). Compared to the model by Ghafoor et al (2023) , the presented approach outperforms for the dimension of valence by r = 0.30 and for arousal by r = 0.47 (see Table 6 ). The performance is, however, hard to compare with the model by Ghafoor and colleagues as they evaluated their model on a self-created unpublished dataset.…”

“… Ghafoor et al (2023) developed a context-aware emotion classifier called TERMS using a Gaussian mixture model for soft assigning emotional perceptions in a valence-arousal space. The classifier maps emotional perspectives as distributions into a multidimensional space based on single texts.…”

UniC: a Dataset for Emotion Analysis of Videos with Multimodal and Unimodal Labels

“…Scholars in natural language processing (NLP) specifically focus on the modality of text. Their research ranges from investigating and modelling the overall sentiment or polarity in a given text [9,10] to the fine-grained detection of specific polarities attached to different aspects of a product or event [11][12][13], emotion detection in text from tweets or other sources [14,15], and other emotion-related topics such as humour and irony [16][17][18].…”

Exploring Emotion Analysis Using Artificial Intelligence, Geospatial Information Systems, and Extended Reality for Urban Services

“…With the rise of social networks, big data have been generated through user opinions on various platforms-such as Facebook and Twitter-the feelings of users being entered into these sources via texts, making these sources and their analyses crucial for learning about user emotions and behavior. Different models have been used to solve this problem, including deterministic models with continuous fine-grained alternatives for affective text analysis, and dimensional models with supervised or unsupervised classification approaches-such as deep-learning techniques and machinelearning techniques [18]. In other words, we can group textual emotion recognition methods into four categories-that is, keyword-based, lexicon-based, machine-learning-based, and hybrid methods.…”