aiqbench analysis
This research challenges the foundational assumption in Emotion Recognition in Conversations (ERC) that every utterance requires a single, unambiguous emotion label. By leveraging Large Language Models as judges and conducting human re-annotation studies, the authors demonstrate that many perceived model errors actually stem from inherent ambiguity in conversational data rather than poor model performance. Consequently, standard single-label evaluation metrics are insufficient for accurately assessing current capabilities in empathetic AI applications.
Exposing Weaknesses in Emotion Recognition in Conversations
arXiv:2609.05806v1 Announce Type: new Abstract: Emotion Recognition in Conversations (ERC) aims to identify speakers' emotions in multi-turn dialogue. Accurate emotion recognition can support a wide range of applications, including empathetic conversational agents, mental health support, and educational technologies. While many recent approaches rely on task-specific fine-tuning, such models may exploit dataset-specific cues. A central yet rarely questioned assumption in ERC is that each utterance can be assigned a single unambiguous emotion label. To investigate this assumption, we study ERC using Large Language Models (LLMs) in a zero-shot setting while incorporating preceding conversational turns as context. We show that aggregate metrics mask systematic failures. Errors concentrate around utterances containing negations, exclamations, and interjections. This pattern is consistent across all evaluated models, suggesting limitations in the benchmarks rather than model-specific weaknesses. A controlled re-annotation study involving four human annotators supports this finding: strong agreement is observed in only 35 percent of cases, with neutral utterances dominating high-agreement instances, while many emotional categories fall into low-agreement regimes. These findings suggest that many apparent model errors reflect genuine annotation ambiguity rather than poor emotion understanding. Standard single-label evaluation is therefore insufficient. To address this limitation, we introduce an LLM-as-Judge framework that evaluates each emotion independently according to its plausibility in the conversational context rather than enforcing a single-label decision.
Abstract:Emotion Recognition in Conversations (ERC) aims to identify speakers' emotions in multi-turn dialogue. Accurate emotion recognition can support a wide range of applications, including empathetic conversational agents, mental health support, and educational technologies. While many recent approaches rely on task-specific fine-tuning, such models may exploit dataset-specific cues. A central yet rarely questioned assumption in ERC is that each utterance can be assigned a single unambiguous emotion label. To investigate this assumption, we study ERC using Large Language Models (LLMs) in a zero-shot setting while incorporating preceding conversational turns as context. We show that aggregate metrics mask systematic failures. Errors concentrate around utterances containing negations, exclamations, and interjections. This pattern is consistent across all evaluated models, suggesting limitations in the benchmarks rather than model-specific weaknesses. A controlled re-annotation study involving four human annotators supports this finding: strong agreement is observed in only 35 percent of cases, with neutral utterances dominating high-agreement instances, while many emotional categories fall into low-agreement regimes. These findings suggest that many apparent model errors reflect genuine annotation ambiguity rather than poor emotion understanding. Standard single-label evaluation is therefore insufficient. To address this limitation, we introduce an LLM-as-Judge framework that evaluates each emotion independently according to its plausibility in the conversational context rather than enforcing a single-label decision.
Artificial Intelligence (cs.AI)
arXiv:2609.05806 [cs.AI]
(or arXiv:2609.05806v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.05806 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 304 words
Related coverage
Employee Burnout: Signs, Causes, and How to Prevent It
Slack Blog · Sep 23, 2026
Spider - Man : Brand New Day Wearable AI Cannot Exist Yet : Real Power - Budget Wall Stands in Its Way
techtimes.com · Aug 3, 2026
Open Distillation of Hereditary Traits
AI Alignment Forum · Jul 14, 2026
The Effect of Emotional Context on Large Language Models' Endorsement of Premature Decisions: Comparing Emotional Vulnerability Across Six Commercial Models
arXiv · Sep 1, 2026