Neural network-based methods, especially deep learning, have been very successful in tackling the expanding data volume as we move into a digital age. Today, these methods not only are used for low-level cognitive tasks, such as recognizing objects and spotting keywords, but also have been deployed in various industrial information systems to assist high-level decision-making in finance, education, and healthcare. While producing highly accurate predictions on datasets, those artifacts provide little understanding of the internal features and representations of the data. Although much effort has been devoted to opening the black-box of neural networks, e.g., sensitivity analysis, the interpretability problem generally worsens as the model complexity grows.
The potentially broad societal impacts of neural network-based methods alert people to a dystopian future and re-ignite research on neuro-symbolic AI: a key idea to mitigate unexpected model behavior and inject interpretability by combining learnable parameters (neuro-) with predefined knowledge templates (symbolic). Recent initiatives involve both the academia and company players, such as IBM and DeepMind.
We have seen various types of logic parameterized and applied to such systems, including fuzzy logic, first-order logic, Boolean logic, and probabilistic logic. Instead of learning rules, another way of building neuro-symbolic AI is to leverage existing knowledge bases and to fuse the information at some stage. Both ways have achieved sound improvements in sentiment analysis and deepened our understanding of affective computing and the cognitive root of human emotion.
In this context, this special issue aims to further stimulate discussion on the design, use, and evaluation of neuro-symbolic AI, with an emphasis on human factors and societal implications. We invite theoretical work and review articles on practical use cases of neuro-symbolic AI that discuss sentiment analysis, emotion recognition, and social computing in general. Original works that help mediate and generate insights on human information behaviors, human-system interactions, and affective states with neural network-based models are also encouraged. Topics of interest include, but are not limited to:
- Neuro-symbolic AI for sentiment and emotion analysis in social media
- Linguistic knowledge in deep neural networks for sentiment analysis
- Integrating knowledge for opinion mining
- Aspect-based, multimodal, and multilingual aspects of sentiment analysis
- Critical assessments of existing sentiment analysis methods
- Explainable sentiment and emotion predictions
- Theoretical foundations of neuro-symbolic AI for affective computing
- Commonsense reasoning for sentiment analysis
- Semantic models for affective computing
- Phrase structure grammar for sentiment analysis
- Conversational sentiment analysis
- Joint sentiment analysis and sarcasm/irony detection
- Sentiment analysis and language learning theory
- Sentiment analysis and social network analysis
- Sentiment analysis and stress/suicide detection
- Sentiment analysis and forecasting methods
- Submission Deadline: 31 January 2022
- Peer Review Due: 1 May 2022
- Revision Due: 15 July 2022
- Final Decision: 1 September 2022
- Publication: September 2022
For author information and guidelines on submission criteria, please visit the TAC Author Information page. Please submit papers through the ScholarOne system, and be sure to select the special-issue name. Manuscripts should not be published or currently submitted for publication elsewhere. Please submit only full papers intended for review, not abstracts, to the ScholarOne portal. Abstracts should be sent by email to the guest editors directly.
Frank Xing, National University of Singapore, Singapore
Iti Chaturvedi, James Cook University, Australia
Erik Cambria, Nanyang Technological University, Singapore
Amir Hussain, Edinburgh Napier University, UK
Björn Schuller, audEERING GmbH, Germany