Emotion recognition in cross-linguistic legal context

The datasets comprise 88 participants' concurrent electrodermal activity (EDA) and heart rate variability (HRV) recordings collected during viewing 20 video clips. The HRV features computed in 120‑s windows for RMSSD, SDNN, pNN50, LF/HF ratio), normalized across participants. The EDA data sampled at 4 Hz and temporally aligned to the task structure, with engineered EDA features extracted from the continuous signals. The repository includes the datasets generated and analyzed for this study; code and inputs for the PANAS logistic-regression analysis and cross-tool agreement analyses; and an executable framework documenting the physiological preprocessing workflow, LSTM architectures, input and fusion structures, training configurations, and software dependencies. Sample data and a complete tensor schema are provided to demonstrate preprocessing, participant-grouped cross-validation, model construction, and short training. GPT-4o and DeepSeek-R1 were queried through their public interactive interfaces rather than through an API; therefore, the repository provides the exact prompts and operating procedure, together with the submitted transcripts, raw model outputs, and manual coding sheets. Detailed requirements and execution instructions are included in the repository.