Sentiment Analysis as a Multidisciplinary Research Area

Erik Cambria, Nanyang Technological University, Singapore 
Frank Xing, Nanyang Technological University, Singapore
Mike Thelwall, University of Wolverhampton, UK
Roy Welsch, MIT Sloan School of Management, USA

IEEE Transactions on Artificial Intelligence


Sentiment analysis is an important natural language processing (NLP) task in its own right and as part of text processing systems. It is widely used to process and understand Internet texts, which is important for both the scientific community and the business world. Its scientific importance is based on many open challenges to define and precisely capture implicit and explicit sentiment. While most works approach it as a simple categorization problem, sentiment analysis is actually a complex research problem that requires many other NLP tasks, including subjectivity detection, anaphora resolution, word sense disambiguation, sarcasm detection, and aspect extraction. Its practical importance is due to the remarkable benefits to be had from some proven commercial applications, including but not limited to user profiling, digital marketing, and financial prediction.

Sentiment analysis has advanced with progress in the construction of knowledge bases for the identification of polarity in text, e.g., SentiWordNet, SentiStrength, and SenticNet, as well as progress in statistics-based approaches that leverage sentiment expression data sets, i.e., machine learning and deep learning. Despite the expanding size of knowledge bases and data sets, and the complicating algorithms, there is a declining trend in linking sentiment analysis with theories from other disciplines. This, from a scientific perspective means considering contributions from linguistics, psychology, and cognitive science, and from a practical perspective means producing task-aware results for downstream applications.

As an area artificial intelligence (AI), sentiment analysis faces similar challenges in requiring broad interdisciplinary input, explainability, etc. Therefore, the research should not only “go deep”, but also “go broad”. We envision that introducing new theories would also mitigate the interpretability problem, i.e., the problem that sentiment analysis methods become data-dependent and function in a black-box manner. Having many aspects of the same problem may help stepping forward in the path from NLP to natural language understanding. Another way to “go broad” would be the combination with other applications, e.g., social computing and recommender systems, where sentiment analysis may yield unexpected results or new insights.

This special issue focuses on bringing multidisciplinary knowledge into sentiment analysis. We expect submissions that introduce theories not usually part of the standard sentiment analysis framework, and potentially attract researchers to learn more about the relevant literature. Minor improvements, e.g., a new neural network architecture that changes performance but lacks a rationale, and applications of the same method on a different domain or dataset fall outside the scope of this special issue.

This special issue focuses on emerging techniques and trendy applications of sentiment analysis as a multidisciplinary research area. Given the focus of the journal, we expect to receive works that propose new AI algorithms for the advancement of sentiment analysis research. While other disciplines, e.g., semiotics, psychology, linguistics, are surely welcome, the AI component must be there and it must be in line with the state of the art. Mostly, we expect to receive works on textual sentiment analysis, but papers on multimodal sentiment analysis will also be considered. The topics of this special issue include but are not limited to:
- Critical assessments of existing sentiment analysis methods
- Explainable sentiment predictions
- Sentiment of multiword expressions
- Hybrid symbolic and sub-symbolic AI for sentiment analysis
- Theoretical foundations of AI for sentiment analysis
- SenticNet 6 and other hybrid knowledge bases for sentiment analysis
- Sentic LSTM and other hybrid deep nets for sentiment analysis
- Commonsense reasoning for sentiment analysis
- Semantic models for sentiment analysis
- Phrase structure grammar for sentiment analysis
- Conversational sentiment analysis
- Morphological hints for sentiment analysis
- Joint sentiment analysis and sarcasm/irony detection
- Sentiment analysis applications
- Sentiment analysis and language learning theory
- Sentiment analysis and social network analysis
- Sentiment analysis and stress/suicide detection
- Sentiment analysis and forecasting methods

Paper submission: 31st March 2021
Initial review feedback: 15th May 2021
Revision: 7th June 2021
Second review feedback: 2nd August 2021
Expected publication date: October 2021