Artificial intelligence (AI) has the potential to be an effective instrument in detecting questionable open-access journals by examining website attributes such as layout and content, based on recent studies. The goal of researchers was to evaluate AI’s capability to replicate human discernment in identifying dubious journals and revealing essential predictive indicators. These ‘dubious’ journals were characterized as those that do not comply with the best practices established by the Directory of Open Access Journals (DOAJ).
The AI model converted journal websites into machine-readable data according to DOAJ standards, including editorial board qualifications and publication ethics. For the classifier of questionable journals, the researchers utilized a dataset consisting of around 12,800 approved and 2,500 non-approved journals and extracted features from the content, design, and bibliometric data of the sites.
When tested on over 15,000 open-access journals sourced from the Unpaywall database, the model identified 1,437 potentially dubious journals, with 1,092 likely to be authentically questionable. These journals contained thousands of articles, attracted millions of citations, received funding from significant organizations, and were prone to engaging authors from developing nations.
Approximately 345 cases were marked as false positives, frequently due to inaccessible sites, formally ceased publications, or misunderstanding with book series or conferences sharing similar titles. Researchers approximated that around 1,780 problematic journals might still go unnoticed.
The study determined that AI could proficiently pinpoint questionable journals, aligning closely with expert evaluations. Nonetheless, they highlighted the necessity for ongoing updates to AI models to adjust to evolving trends. Future advancements should investigate real-time web crawling and community input to refine AI-based tools, nurturing a dynamic system to uphold research integrity.