Correcting Citation Inflation via Bayesian Knowledge Integration: A Scalable Framework Validated on OpenAlex Data

This paper introduces the Bayesian Knowledge Integration Index (BKII), a novel bibliometric metric designed to evaluate the quality and interdisciplinary integration of scientific contributions while correcting for citation inflation and manipulative practices. BKII employs a hierarchical Bayesian model using Negative Binomial likelihood to handle overdispersed citation counts, integrated with Rao-Stirling diversity, entropy-based novelty measures, and NLP-derived functional citation weights. We validate BKII empirically on a representative sample of 10,000 publications from OpenAlex, demonstrating its superior ability to predict prestigious awards such as Nobel Prizes and Fields Medals with an ROC AUC of 0.89, compared to 0.80 for the h-index. Retrospective analyses on 50 Nobel laureates in Physics versus 50 matched non-laureates confirm BKII's enhanced discrimination. Through Monte Carlo simulations, agent-based modeling, sensitivity analyses, and comparisons with established metrics (h-index, g-index, i10-index, h5-index, CNCI, m-index, disruption index, innovation index, EDM, RCR) and MCMC implementations (Stan, PyMC, JAGS), we demonstrate BKII's superior robustness and predictive accuracy under various stress scenarios. The framework provides credible intervals, anomaly detection for citation cartels, and field-normalized scores, offering a reproducible tool for academic evaluations, funding decisions, and policy-making. Code and data processing scripts are available on GitHub for community verification. No external funding was received.