Empirical analysis of Zipf's law, power law, and lognormal distributions in medical discharge reports.

摘要: Abstract Background Bayesian modelling and statistical text analysis rely on informed probability priors to encourage good solutions. Objective This paper empirically analyses whether in medical discharge reports follow Zipf’s law, a commonly assumed property of language where word frequency follows discrete power-law distribution. Method We examined 20,000 from the MIMIC-III dataset. Methods included splitting into tokens, counting token frequency, fitting distributions data, testing alternative distributions—lognormal, exponential, stretched truncated power-law—provided superior fits data. Result Discharge are best fit by lognormal distributions. appear be near-Zipfian having provide over pure power-law. Conclusion Our findings suggest that report would benefit using non-parametric models capture behavior.