A data mining-based study on academic publication retractions in the 21st Century

<b>Background:</b> The rising number of academic retractions has drawn increasing attention across the academic community. With the availability of large-scale retraction data from Crossref and Retraction Watch, systematic analysis of academic retractions has become feasible.<b>Methods:</b> This study examines all retracted academic publications from the 21st century up to June 4th, 2025. By using BERTopic, Apriori, and data visualization techniques, we've conducted a comprehensive analysis across six subjects with over 6,000 retractions of each subject.<b>Results and conclusions:</b> Our findings detail retraction counts, durations, topic trends, author nationalities, publishers, retraction reasons, and associations among these factors. The overall number of retractions has been continuously rising, with sharp increases in 2010 and 2020 to 2023, and the peak occurring in 2023. The primary reasons for retractions in biomedical studies are paper mills and issues with data and images, with third parties being the main initiators of investigations. In computer science and technology, retractions are mainly due to referencing and attribution issues, as well as unreliable results, with journals, conferences, and publishers often initiating the investigations. We also offer some suggestions that can help monitor research misconduct in academic publications.