Protocol-bounded semantic alignment of pain catastrophizing and traditional Korean medicine pattern constructs: a computational literature-based discovery study

This repository contains the supporting data, audit files, knowledge-graph outputs, semantic-alignment outputs, and protocol/model-lock-in materials for the manuscript: “Protocol-bounded semantic alignment of pain catastrophizing and traditional Korean medicine pattern constructs: a computational literature-based discovery study.” The repository supports a computational literature-based discovery study designed to generate auditable, hypothesis-generating alignment signals between Pain Catastrophizing Scale subdomains and traditional Korean medicine pattern constructs. The study does not include individual participant data, patient-level clinical data, treatment-response data, or human-subject intervention data. The deposited files include: (1) the final corpus assignment table; (2) the entity dictionary; (3) the relation audit table; (4) knowledge-graph node, edge, evidence, metric, path, and JSON graph files; (5) final semantic alignment scores, top-ranked candidate pairs, robustness summaries, rank-stability outputs, and negative-control outputs; and (6) protocol and model-lock-in files, including the screening codebook, protocol amendment log, environment specification, AI-use log, and semantic model lock-in JSON. The final analytic scope is restricted to the author-approved PubMed Central-retrieved full-text corpus. Records outside the final manuscript scope were not treated as negative evidence and were not adjudicated as full-text exclusions. No copyrighted full-text articles or full article bodies are redistributed in this repository. The repository contains derived metadata, record identifiers, entity labels, audit decisions, graph structures, semantic scores, validation outputs, and protocol materials needed to interpret and reproduce the manuscript results within the stated analytic boundary. The primary semantic embedding model used in the final locked analysis was BAAI/bge-m3 in dense-embedding mode only, with Hugging Face revision SHA 5617a9f61b028005a4858fdac845db406aefb181. Robustness analyses used intfloat/multilingual-e5-large and nlpai-lab/KURE-v1. The outputs are intended to support transparency, reproducibility, and independent scrutiny of the manuscript’s computational workflow. They do not establish diagnostic equivalence, clinical validity, causality, treatment efficacy, or direct construct identity between pain catastrophizing subdomains and traditional Korean medicine pattern constructs.