Who did what? A TF-IDF-based contribution metric for team science

As science becomes increasingly team-based, mapping scientific contributions has become essential for understanding team dynamics and strengthening scientific accountability. While existing frequency-based measures have offered valuable insights, they may not fully capture heterogeneity in individual specialization and shared responsibilities. Complementing conventional approaches, this study proposes a new TF-IDF-based framework that characterizes author contribution profiles by reweighting raw role counts according to both the diversity of roles an author performs and the breadth of role distribution across co-authors. The study empirically implements and demonstrates the advantage of this method using a large corpus of publication metadata from PLOS ONE. Results show that the TF-IDF-based framework filters out uninformative “all-hands” contributions, captures contribution specificity that is otherwise masked by raw role counts, and sharpens the differentiation among author contribution profiles. In predicting authorship positions using random forest classifiers, TF-IDF-based contribution features consistently outperform raw count features, and this advantage holds across team sizes and research domains. Additional analyses show that TF-IDF-based measures offer the greatest advantage over raw role counts in moderately sized, relatively equal teams, and among highly specialized individuals. Overall, the TF-IDF-based framework provides a practical tool for a finer-grained, more differentiated account of labor division within research teams, offering a unique way to enhance transparency in authorship attribution.