ViTOED: A Dataset for Target-Oriented Emotion Detection on Vietnamese Social Media Texts

2026-08-12

AI Summary

This study addresses the scarcity of high-quality annotated data for target-oriented emotion detection in Vietnamese social media by introducing ViTOED, the first benchmark dataset comprising 10,985 comments and 21,244 quadruple annotations (source, target, emotion expression, and polarity). The authors develop a structured sentiment graph framework to guide rigorous annotation protocols and systematically evaluate various Vietnamese pre-trained language models. Their analysis uncovers significant linguistic challenges, including implicit agents and lexical ambiguity, and demonstrates that current models exhibit substantial limitations in span identification and relation extraction. This work not only establishes the first fine-grained emotion analysis benchmark for Vietnamese but also provides critical insights and directions for future model improvement.

Abstract

This paper introduces ViTOED, a novel dataset for target-oriented emotion detection in Vietnamese social media texts. The ViTOED comprises 10,985 user comments and 21,244 manually annotated opinion quadruples (source, target, expression, polarity) that follow strict guidelines. The dataset reveals Vietnamese-specific phenomena, such as implicit sources and targets and vocabulary ambiguities, enabling deeper analysis of user emotions toward entities. We propose a baseline using structured sentiment graphs and evaluate various Vietnamese pre-trained language models. The empirical results highlight challenges in span detection and relation extraction and indicate substantial room for model improvement in Vietnamese Target-Oriented Emotion Detection tasks.

Problem

• target-oriented emotion detection
• Vietnamese social media
• opinion quadruple
• implicit source and target
• sentiment analysis

Innovation

• target-oriented emotion detection
• structured sentiment graph
• Vietnamese social media
• opinion quadruple annotation
• pre-trained language models