This paper aims to analyse the relationship between two measures of university quality, the outcome and other characteristics of a mandatory accreditation and the university position in the national ranking.
Natural language processing (NLP) models are used to calculate the sentiment indicators for 1,850 accreditation reports from the Polish Accreditation Agency. The sentiment indicators, accreditation frequency and outcomes for 203 HEIs are used in correlation analysis, automated linear regressions and quantile regressions with the university position in the Polish Perspektywy rankings as the outcome variable.
High/low frequency of accreditation visits, excellent/poor accreditation outcomes and low/high frequency of negative inclination words in the accreditation report are followed by high/low university rankings. Quantile regressions reveal that these relationships vary with the quality of the university.
Publishers of university rankings may consider adding the accreditation features to the set of indicators used in such rankings. The machine learning methodology presented allows cross-country inconsistencies to be identified in the approaches used by accreditation agencies in Europe. The authors of the accreditation reports should be aware they can be mined by machine learning models and this should be considered when the reports are drafted.
This is a novel application of NLP models for analysing the relationship between the accreditation and rankings of universities. In other research, the author has applied NLP models to test whether quality assurance agency (QAA) accreditation in the UK can predict how students rate their university on whatuni.com website.
