Long-horizon Naturalistic Interaction
168 semi-structured interviews conducted by licensed counselors, eliciting authentic nonverbal cues across extended social interactions, not thin-slice impressions.
Clinically Grounded Personality Computing
Understanding personality from visual behavior remains challenging. Existing benchmarks rely on short, crowd-annotated clips and capture only coarse impressions, lacking temporal continuity, clinical validity, and fine-grained sub-trait structure.
Expert-validated BFI-2 Labels
Continuous scores for all five Big Five domains and, for the first time in PTI benchmarks, fifteen validated BFI-2 sub-traits, scored independently by two psychology professionals.
Research Dataset Access
A 5,000-clip research subset with expert BFI-2 annotations is now available on Hugging Face through individual application and manual approval.
The PersoMoni Benchmark
PersoMoni bridges the gap between ecological validity and computational tractability, preserving the richness of full interviews while providing localized clip units for learning.
Current release on Hugging Face
A distribution-aware subset of 5,000 processed facial clips is available now: 3,997 training clips, 1,003 test clips, 199,897 JPEG frames, and 20 BFI-2 targets. Access is granted to named researchers after application and manual review. The current release does not include the complete original interview videos. Full-dataset feature-map processing and quality control remain in progress.
Collection Protocol
- Semi-structured interviews with licensed psychological counselors on self-concept, relationships, and values
- Pre-interview screening via SCL-90 and BFI-2 to tailor questioning
- 1080p front-facing recording at 30 fps, seated natural posture, continuous eye contact
- Face-centered processing via YOLO detection, temporal segmentation into clip units
- Three-stage QC: individual scoring, cross-review, and consensus discussion
Key Properties
| Participants | 168 (86 F / 82 M) |
|---|---|
| Video clips | 20,000+ face-aligned segments |
| Duration | ~25 min avg. per interview |
| Resolution | 1080p @ 30 fps |
| Label space | 20-dim continuous BFI-2 vector |
| Annotation | Dual expert rating + averaging |
| Demographics | Balanced age & education groups |
From 5 Traits to 15 Sub-traits
PersoMoni is the first personality computing benchmark to extend granularity from 5 to 15 dimensions, aligned with the psychometrically validated BFI-2 hierarchical framework.
Extraversion
- Sociability
- Assertiveness
- Energy Level
Agreeableness
- Compassion
- Respectfulness
- Trust
Conscientiousness
- Organization
- Productiveness
- Responsibility
Negative Emotionality
- Anxiety
- Depression
- Emotional Volatility
Open-Mindedness
- Intellectual Curiosity
- Aesthetic Sensitivity
- Creative Imagination
How to Cite
If you use the PersoMoni dataset in your research, please cite our paper.
@article{cui2026persomoni,
author = {Cui, Feng-Qi and Huang, Jinyang and Zhao, Sirui and Li, Kun and Liu, Zhi and Li, Meng and Jia, Ziyu and Guo, Dan and Wang, Meng},
journal = {IEEE Transactions on Affective Computing},
title = {PersoMoni: A Comprehensive Video-Based Benchmark Dataset for Fine-grained Personality Assessment with 15 Trait Dimensions},
year = {2026},
pages = {1--14},
doi = {10.1109/TAFFC.2026.3698795},
url = {https://ieeexplore.ieee.org/document/11543185}
}