Sentiment Simulation using Generative AI Agents

The preprint challenges the existing paradigm of sentiment analysis as retrospective classification based on surface-level linguistic features. It proposes a shift towards prospective and dynamic sentiment simulation grounded in psychological principles, utilizing generative AI agents embodied with psychographic profiles.

This preprint presents a compelling case for a Model Revolution. The authors argue for a fundamental shift in how sentiment analysis is conducted, moving from simply classifying sentiment based on text to simulating sentiment responses by embedding psychological profiles within AI agents. They demonstrate the effectiveness of their approach by showing that these agents can replicate real survey results with high fidelity and generate nuanced and contextually appropriate responses to various socio-political and economic scenarios. While stopping short of proposing a complete theoretical overhaul of sentiment analysis, the combination of generative AI and behavioral science to simulate sentiment represents a significant departure from current practices and has the potential to reshape the field if widely adopted. There are elements of a potential Paradigm Shift here, but the work presented appears to still be within the Model Revolution phase given further validation and widespread adoption in the future is still needed.