AI second guess that emotion

Research in the International Journal of Computational Intelligence Studies has looked at how we might improve artificial intelligence (AI) systems for interpreting human emotion in written communication. The new system is capable of identifying sentiment not only in broad terms, positive, negative, and neutral, but also at a more detailed, aspect-specific level.

Sentiment analysis usually evaluates entire sentences or documents as a single unit. This can hide the subtleties of human expression. For instance, a restaurant review may praise the food while criticising the service. Previous AI models could struggle to separate these differing opinions, often assigning a generalised sentiment score. The new model overcomes this limitation by emphasising emotionally charged keywords, the words that carry the most significant emotional weight in a sentence. It does this using an attention network, a computational mechanism that allows AI to prioritise certain inputs over others.

This focus on the most emotional terms in a piece of text allows the AI to classify sentiment directed at specific aspects of a text. In the restaurant example, the model can distinguish the positive sentiment aimed at the food from the negative sentiment about the service, producing a more nuanced interpretation. Moreover, the system’s ability to pay attention to the most emotionally charged words is a useful advance in natural language processing.

Such a tool could help businesses that rely on customer feedback, social media analysis, and online reviews. With it a company could spot concerns being discussed online as they arise and so make a timely response to help manage their image and refine their marketing. They might even be able to offer targeted responses to individuals or groups to improve customer satisfaction and perception.

This research is part of a growing trend in AI research towards improving the way in which computers interpret language and emotion. By enabling machines to analyse sentiment at the level of individual aspects rather than entire texts, this approach contributes to the development of more perceptive, context-aware AI.

Yuan, Z. and Yuan, J. (2026) ‘Aspect-level sentiment classification with emotional keywords attention network’, Int. J. Computational Intelligence Studies, Vol. 13, No. 5, pp.1–13.