Manuscript Classification to Support the Analysis of Biases in Publication Opportunities

Diversity in the research workforce is essential for addressing complex problems. Female investigators, along with individuals from diverse backgrounds and identities, generate novel, impactful, and innovative research, whereas biases may affect their chances of publication. In this article, we present ad hoc strategies for classifying biomedical manuscripts according to four dimensions (experimental designs, medical specialties, sample sizes and funding sources). This classification is intended to further explore how gender bias may affect publication acceptance by adjusting the analyses for these important confounders. 56 manuscripts were manually annotated by two experts to assess the performance of the models. The proposed strategies performed well (compared to the inter-annotator agreement) with accuracy scores of 75%, 78%, 78% and 82%, for the above-mentioned dimensions, respectively. Given the timely importance of such questions in research, we are planning to apply the categorization at the scale of MEDLINE.