Enhancing Emotion Recognition in Conversations through Global Context: An Empirical Analysis
Abstract: According to multimodal and contextualized nature of the human conversation, correctly identifying an emotion for given utterance in the conversation has always been a challenging task. Recent research benefits from Graph Neural Networks by capturing implicit relationship of temporally proximate utterances. In this paper, we expand the structure of the graph exploited by these models reflecting the global context of the conversation and explore how leveraging conversational context and interactions can lead to more accurate emotion recognition. We empirically analyze the modules on Emotion Recognition in Conversation models, showing this approach enhances the performance of these models. Our experiments show that incorporating global conversational context has a positive effect on the performance of emotion recognition.
Paper Type: Long
Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining
Research Area Keywords: Multimodal Emotion Recognition in Conversation, Graph Neural Network
Contribution Types: Model analysis & interpretability
Languages Studied: English
Submission Number: 5834
Loading