Beyond Emotion Scores: Reproducibility Data and Validation Materials for a Systematic Review of Emotional AI (2019–2026)
Abstract
This dataset provides the supplementary data and reproducibility materials accompanying the systematic literature review Beyond Emotion Scores: Evaluating Emotional AI Across Benchmarks, Cultures, and Time in the LLM Era - A Systematic Review. The review covers literature published from 1 January 2019 to 26 September 2026 and examines benchmark and construct validity, cultural robustness, temporal affect modelling, and emotional capability evaluation in large language models and multimodal foundation models. The reproducibility materials include the search protocol and provenance records, a 130-record screening ledger, documented eligibility decisions, nine full-text exclusions, the characteristics and thematic coding of 93 included studies, study-level Q1–Q5 quality appraisal, reference-integrity records, a fixed 20-record independent human screening exercise, supplementary external-LLM validation materials, and scripts for regenerating the principal corpus statistics and human screening agreement results. The Affective-AI Evidence-to-Governance Matrix (AEGM) developed in the accompanying manuscript connects these findings to evaluation and governance requirements. The repository contains supporting research documentation and analysis data, not a dataset of individual human participants.
