The Super Emotion Dataset

Enric Junqué de Fortuny

cs.CLMay 21, 2025

The Super Emotion Dataset

Authors: Enric Junqué de Fortuny

Organizations: Managerial Decision Making IESE, Barcelona, Spain

Abstract

Despite the wide-scale usage and development of emotion classification datasets in NLP, the field lacks a standardized, large-scale resource that follows a psychologically grounded taxonomy. Existing datasets either use inconsistent emotion categories, suffer from limited sample size, or focus on specific domains. The SuperEmotion dataset addresses this gap by harmonizing diverse text sources into a unified framework based on Shaver's empirically validated emotion taxonomy, enabling more consistent cross-domain emotion recognition research. This report describes version 2 of the dataset. We also report baseline results for a fine-tuned ModernBERT encoder.

Figures & tables

Dataset	Classes	Original release	Train	Val.	Test
MELD	7	13,708	8,901	966	2,256
CARER	6	416,809	318,668	37,723	37,428
GoEmotions	28	54,263	41,840	5,218	5,208
CrowdFlower	13	39,998	31,210	3,896	3,893
SemEval	11	10,690	6,567	866	3,165
SuperEmotion	7	535,468	407,186	48,669	51,950

Table 1: The source datasets: emotion classes, size of the original release, and texts in each SuperEmotion split (version 2, after label mapping and deduplication). We created the splits of CARER and CrowdFlower.

Emotion	MELD	CARER	GoEmotions	CrowdFlower	SemEval
anger	anger, disgust	anger	anger, annoyance, disapproval, disgust	anger, hate	anger, disgust
fear	fear	fear	fear, nervousness	worry	fear
joy	joy	joy	amusement, excitement, joy, optimism, pride, relief	enthusiasm, fun, happiness, relief	joy, optimism
love		love	admiration, approval, caring, desire, gratitude, love	love	love, trust
sadness	sadness	sadness	disappointment, embarrassment, grief, remorse, sadness	sadness	pessimism, sadness
surprise	surprise	surprise	confusion, realization, surprise	surprise	surprise

Table 2: Mapping of the source labels to Shaver’s categories, per source dataset.

Source	neutral	surprise	fear	sadness	joy	anger	love	Dropped	Total
MELD	5,662	1,311	344	951	2,042	1,854	0	0	12,164
CARER	2	14,076	46,284	120,139	138,846	56,449	32,809	0	408,605
GoEmotions	17,723	4,287	926	4,021	7,615	7,818	15,423	2,723	60,536
Crowdflower	8,782	2,182	8,442	5,155	9,237	1,433	3,792	827	39,850
SemEval	0	566	1,848	3,607	5,065	4,779	1,757	1,527	19,149
Total	32,169	22,422	57,844	133,873	162,805	72,333	53,781	5,077	540,304

Table 3: Labels per emotion category and source in SuperEmotion version 2, all splits together. The 41 texts found in two sources are counted with the first.

Emotion Recognition

Emotion recognition uses language, voice, face, body movement, and physiological signals to infer emotional categories or dimensions such as valence and arousal.

See all 129 papers

SpEmoC: A Balanced Speaker-Segment Multimodal Emotion Benchmark
Jul 20, 2026Sania Bano, Shahzad Ahmad, Santosh Kumar Vipparthi +2Emotion Recognition in Conversations Multimodal Emotion Recognition
CVPR Lab, Indian Institute of Technology Ropar, India · Østfold University College, Norway · CVPR Lab, SCSS, Trinity College Dublin, Ireland
EmoS: A High-Fidelity Multimodal Benchmark for Fine-grained Streaming Emotional Understanding
May 9, 2026Pengze Guo, Jingxi Liang, Zhiwen Xie +2Multimodal Large Language Models Multimodal Emotion Recognition
NLP2CT Lab, Department of Computer and Information Science, University of Macau · School of Computer Science, Central China Normal University · University of Macau
Multilingual Multi-Label Emotion Classification at Scale with Synthetic Data
Apr 14, 2026Vadim BorisovEmotion Recognition Multilingual Text Classification

Jul 20, 2026J/K move · Enter open · S save

SpEmoC: A Balanced Speaker-Segment Multimodal Emotion Benchmark

Sania Bano, Shahzad Ahmad, Santosh Kumar Vipparthi, Sukalpa Chanda+1

CVPR Lab, Indian Institute of Technology Ropar, India · Østfold University College, Norway · CVPR Lab, SCSS, Trinity College Dublin, Ireland

Emotion Recognition in Conversations Multimodal Emotion Recognition Cross-Dataset Generalization

Open paper AskPDF arXiv

Understanding human emotions in spoken conversations is a key challenge in affective computing, with applications in empathetic AI, human computer interaction, and mental health monitoring. However, existing datasets vary in scale, emotion distribution, modality alignment, and data partitioning strategies, which can influence reliable cross-dataset generalization and minority-emotion modeling. We introduce SpEmoC a Speaking segment Emotion for Conversations comprising 306,544 raw clips from 3,100 English language movies and TV series. From these, 30,000 high quality, class balanced clips are curated, featuring synchronized visual, audio, and textual modalities annotated for seven emotions through a hybrid pipeline that integrates pretrained models with human validation. SpEmoC uses strict movie- and series-level splits to prevent content overlap between split sets, allowing more reliable evaluation of model generalization. The dataset also maintains a near-balanced distribution across seven emotions, including minority classes such as Fear and Disgust, which supports more balanced learning across categories. Extensive experiments, including in-domain benchmarking, cross-dataset transfer, low-data training, class-imbalance analysis, and modality transfer show that balanced data and careful splitting lead to more stable performance across emotions when models are evaluated on other datasets. These results highlight the importance of dataset design for robust and transferable multimodal emotion recognition.

Abstract

Emotion Recognition

Explore similar work

SpEmoC: A Balanced Speaker-Segment Multimodal Emotion Benchmark

EmoS: A High-Fidelity Multimodal Benchmark for Fine-grained Streaming Emotional Understanding

Multilingual Multi-Label Emotion Classification at Scale with Synthetic Data