Naturalistic Social Mind Neuroimaging Database (NSM)

# Naturalistic Social Mind (NSM): A theory‑informed reference dataset for naturalistic social cognition ## Overview NSM is a BIDS‑formatted (v1.9.0) fMRI dataset combining brain responses from 40 healthy adults (20 females, 24.28 ± 2.79 years) watching eight videos (~35 min) with >150,000 theory‑driven, time‑resolved social annotations (2‑s resolution, matching TR). A 6‑min resting‑state scan is also included. The dataset supports research on naturalistic social cognition, encoding/decoding models, and benchmarking computational models against human judgments. ## Stimuli Eight videos cover positive/negative interactions, dialogue‑present/absent, live‑action/animated formats, plus a low‑social‑content control (Banana Kong). Due to copyright, only metadata and annotations are distributed; original videos are not included. ## Annotation framework Four layers, ten attributes, coded with categorical labels and/or 7‑point dimensional ratings per 2‑s clip: - **Person**: Action (6 dims), State (3 dims), Trait (3 dims) - **Interpersonal**: Relationship (5 dims), Dialogue (transcript) - **Context**: Situation (8 dims), Locale, Object - **Narrative**: Event (semantic summary), Boundary (timestamps) Annotations are provided as CSV files (one per video) and are temporally aligned with fMRI time series. ## MRI acquisition - 3T Siemens MAGNETOM Prisma, 20‑channel head coil - fMRI: TR=2000 ms, TE=30 ms, 58 slices, 2.5 mm isotropic, multiband factor 2 - T1‑weighted MPRAGE: TR=2530 ms, TE=2.27 ms, 1 mm isotropic ## Preprocessing fMRIPrep (v21.0.2) standard pipeline; defaced anatomicals; additional post‑processing for ISC/encoding analyses (6 mm smoothing, nuisance regression). Preprocessed data in `derivatives/`. ## Dataset structure (BIDS) sub-<ID>/anat/ and /func/ (raw + JSON sidecars) derivatives/sub-<ID>/... (preprocessed) stimuli/task-<TASK>/ (metadata + annotation CSVs: person, interpersonal, context, narrative) code/ (analysis scripts) ## Usage notes - Social features co‑vary; use regularized/multivariate methods to handle shared variance. - Annotations are expert consensus, not individual subjective ratings. - No video files – only temporal boundaries and metadata are provided. ## Citation Please cite the associated Data Descriptor (Zhang et al., 2026) and include the dataset DOI when using these data. ## Ethics Approved by IRB of Beijing Normal University (IRB_A_0024_2024002). All identifying information removed; anatomical images defaced. ## Contact Corresponding author: Yin Wang ([e-mail])