The linguistic anatomy of citations: textual and structural predictors in rising research frontiers

Purpose This study aims to examine how linguistic and structural features of research abstracts are associated with citation performance across six selected emerging scientific domains: Blockchain, Climate Science, Gene Editing, Nanomedicine, Quantum Computing and Space Mining. Design/methodology/approach A dataset of 39,350 journal articles (2015–2024) from selected domains was retrieved from the Scopus database and analysed using computational text mining techniques in Python. Six abstract-level features length, readability (Flesch Reading Ease), subjectivity, polarity, sentiment and keyword density were correlated with citation counts using Pearson’s correlation and regression modelling. Findings Abstracts of 150–350 words receive significantly higher citations, with a consistent negative correlation between length and citations (r = −0.13 to −0.18). Keyword density is the strongest bivariate predictor across domains (r = 0.17–0.23), followed by readability (r = 0.08–0.15); subjectivity shows mixed, weak effects, while polarity and sentiment show negligible associations (r = −0.05 to 0.06). Cross-validated multivariate and machine-learning models indicate negligible to negative out-of-sample predictive power for citation counts in every domain. These bivariate correlations should be read as weak, within-sample associations rather than validated predictors. Originality/value This study provides large-scale cross-domain evidence, linking linguistic and structural features of abstracts with citation performance across six emerging research domains, while cautioning, through rigorous cross-validation, that these features alone are weak and have limited generalisability as predictors of citation performance.