PRISM Model: Six-Dimension Psychosocial Classification Dataset

These classifications were generated through theory-guided LLM prompt engineering, in which each psychosocial dimension was extracted using a dedicated prompt explicitly grounded in established psychosocial theory rather than open-ended or ad hoc classification. Important note on scope: this dataset contains only the derived psychosocial classifications, not the original review text. Raw review text is not included because Indeed's terms of service restrict redistribution of user-generated content; sharing only the derived, aggregated classification output avoids this restriction while still supporting reproducibility of the study's analytical findings.

The dataset includes a README sheet describing its structure and provenance, a full data dictionary explaining every column and its theoretical grounding, the classification data itself, and a lookup table of abbreviations used in the metadata fields (e.g., U.S. state codes, employee status codes). This dataset is intended to support verification, reanalysis, and extension of the study's findings by other researchers working in employee voice analytics, organisational text mining, people analytics, and AI-enabled qualitative research methods.

The accompanying analytical code used to generate these classifications is openly available at: github.com/CDAC-lab/employee-review-analysis.