Sentiment (polarity) and emotion are trained and learned jointly in NLP and affective computing

What is it about?

Sentiment analysis (positive vs. negative polarities of the views expressed) and emotion classification (to identify the emotion involved, such as anger, envy happiness, etc.) are two tasks that are typically conducted independently in affective computing and NLP. We propose that these two tasks be performed jointly, leveraging data that are annotated for both sentiment and emotion to improve the performance of both tasks.

Featured Image

Why is it important?

Both sentiment analysis and emotion classification are highly significant but challenging tasks dor for LLM. They are also often the foundation for interpreting language big data to resolve the grand challenges that we face now. One of the bottlenecks to processing either sentiment or emotion using LLM is that the annotation of sentiment and emotion is costly as they cannot be done fully automatically. The proposal to learn sentiment and emotion jointly not only improves the performance, it also opens up the possibility of cross-learning when only either sentiment or emotion information is available.