Persona-E$^2$: A Human-Grounded Dataset for Personality-Shaped Emotional Responses to Textual Events
Summary: arXiv:2604.09162v1 Announce Type: cross
The landscape of affective computing is evolving, and a significant shift is observed in how researchers are approaching the analysis of emotional responses to text. Traditionally, most studies have treated emotion as a static property of text, primarily focusing on the writer’s sentiment. However, this approach has a notable limitation: it often overlooks the reader’s perspective and the influence of individual personality traits on emotional reactions to the same event. This article discusses the introduction of a novel dataset called Persona-E$^2$ (Persona-Event2Emotion), designed to address these challenges.
The Challenge of Emotional Appraisal
In the realm of affective computing, emotional appraisal—the process through which individuals evaluate emotional experiences—is crucial. Yet, research has revealed that existing methodologies often fail to account for the diverse emotional responses elicited by identical events. This shortfall is particularly evident in the context of Large Language Models (LLMs) that aim to replicate nuanced emotional reactions. These models frequently succumb to the “personality illusion,” relying on superficial stereotypes instead of logical and authentic cognitive processes.
Introducing Persona-E$^2$
To bridge the gap between personality traits and emotional responses, the Persona-E$^2$ dataset emerges as a groundbreaking resource. This large-scale dataset is meticulously grounded in the widely recognized Myers-Briggs Type Indicator (MBTI) and Big Five personality traits. It is designed to capture the variations in emotional responses from readers across different contexts, including:
- News articles
- Social media posts
- Life narratives
By employing this dataset, researchers can gain insights into how personality shapes emotional reactions to textual events, thereby enhancing the understanding of emotional dynamics in communication.
Key Findings and Implications
Extensive experiments conducted using the Persona-E$^2$ dataset have yielded significant insights, particularly highlighting the limitations of state-of-the-art LLMs. These models often struggle to accurately capture the nuanced appraisal shifts that are vital for understanding emotional responses, especially within social media contexts. The findings suggest that:
- Incorporating personality information markedly improves comprehension of emotional responses.
- The Big Five personality traits play a crucial role in alleviating the “personality illusion” experienced by LLMs.
This research underscores the importance of integrating personality considerations into affective computing to foster a more nuanced understanding of emotional responses across different textual mediums.
Conclusion
In conclusion, the Persona-E$^2$ dataset represents a significant advancement in the field of affective computing. By grounding emotional responses in human personality traits, it not only enhances the accuracy of emotional appraisal models but also paves the way for more empathetic and personalized AI systems. As the field continues to evolve, the insights gained from this research may lead to a deeper understanding of human emotions and improve the interaction between humans and AI.