A chinese multi-label affective computing dataset integrating mbti personality traits and emotion intensity from social media posts

CMACD: Chinese Multi-label Affective Computing Dataset # 📌 Overview CMACD is a large-scale Chinese affective computing dataset that integrates personality traits (MBTI) with multi-label emotion annotations and intensity scores. The dataset is constructed from social media text and provides a resource for studying the relationship between personality and emotional expression in natural language. --- # 📊 Dataset Statistics Users : 11,336 Posts : 566,800 Posts per user : 50 Text length : 30–150 Chinese characters Personality labels : 16 MBTI types Emotion labels : 6 categories anger, fear, happiness, neutrality, sadness, surprise Annotation type : Multi-label with intensity scores (0–1) --- # 🧠 Annotation Method Emotion annotations are generated using the Extended Quantification Network (EQN) framework based on a BERT model. The model assigns continuous intensity scores to each emotion, enabling both macro- and micro-emotion analysis. For more details, please refer to our paper: *Expansion quantization network: A micro-emotion detection and annotation framework* [https://doi.org/10.1371/journal.pone.0333930](https://doi.org/10.1371/journal.pone.0333930) Feel free to cite our article: Zhou, J., Luo, S., & Chen, H. (2025). Expansion quantization network: A micro-emotion detection and annotation framework. PloS one, 20(11), e0333930. --- # 📁 Data Format The folder CMACD contains 16 subfolders. Each subfolder is named after a personality type and contains several CSV files, with each CSV file corresponding to 50 posts from a single user. The dataset is organized as structured text samples: | text | anger | fear | happiness | neutral | sadness | surprise | | ------------ | ----- | ---- | --------- | ------- | ------- | -------- | | example text | 0.01 | 0.00 | 0.85 | 0.10 | 0.00 | 0.04 | ⚠️ Note: No user identifiers are included No raw profile-level data are provided --- # 🔒 Data Access (Important) Due to privacy and ethical considerations, the dataset is not publicly released in raw form . Researchers may request access for academic purposes only by contacting the corresponding author. Access will be granted under the following conditions: Non-commercial use only No attempt to re-identify users Compliance with research ethics guidelines --- # ⚖️ Ethics Statement All data were collected from publicly accessible sources. All identifiable information has been removed or anonymized. The dataset is shared in a de-identified format and is intended strictly for academic research. --- # 🧪 Benchmark Results We provide baseline results using multiple models: ## Personality Classification (Accuracy) | Model | I/E | N/S | T/F | J/P | | ----- | ---- | ---- | ---- | ---- | | BERT | 0.76 | 0.78 | 0.79 | 0.73 | ## Emotion Classification | Model | E1_Acc | Ex_Acc | | ----- | ------ | ------ | | BERT | 0.92 | 0.88 | --- # 📈 Potential Applications Affective computing Personality-aware NLP Computational psychology Social media analysis Human-computer interaction --- # 📜 License This dataset is available for academic research only . Commercial use is strictly prohibited. --- # 📖 Citation If you use this dataset, please cite: Zhou, J., Luo, S., & Chen, H. (2025). Expansion quantization network: A micro-emotion detection and annotation framework. PloS one, 20(11), e0333930. # ⭐ Acknowledgements We thank the contributors and the research community for supporting open research in affective computing.