Enhancing research integrity through advanced technologies and expert oversight

Over the last few years, fraud has increased across industries, and advancements in technology have added another layer of complexity. While scientific publishing has also seen an increase in fraud, innovations in technology have significantly improved our ability to investigate these instances, making it easier to identify errors and malpractice at scale.

We can now detect fraud and misconduct more effectively than ever thanks to high levels of investment in tools and human expertise. We have improved our ability to detect, correct and remove problematic content at scale, safeguarding the integrity of the research we publish. New technology also helps us prevent the publication of fraudulent research, often stopping problematic papers from entering peer review, thus improving the effectiveness and integrity of the publication process while saving valuable time for our editors and reviewers.

Our responsibility as a publisher to safeguard the scientific record — combined with our improved ability to detect errors and malpractice at scale — has led to an increase in corrections to the record, including errata (correction of errors by the publisher), corrigenda (correction of errors by the author), expressions of concern, article withdrawal, article removal and article retraction. These corrections — including retractions — are a valuable part of the scientific process, and we are committed to carrying them out as transparently and responsibly as possible.

Here are some ways our tools and processes have evolved over the years.

From investigating individual articles to analyzing complex networks

Our tools have evolved from investigating isolated papers to scrutinizing millions simultaneously. We can now screen papers as they are submitted, retrospectively audit the published literature, and uncover intricate networks of misconduct spanning many articles across multiple journals. Alongside our increased ability to detect complex networks, we make sure authors have a chance to respond to concerns.

Adding signals

The number of data points we monitor that indicate potential ethical misconduct has grown consistently over the last few years. Our Research Integrity & Publishing Ethics team has identified and will continue to vet more than 100 research integrity signals across the areas of content and author integrity, person integrity, peer review integrity, editorial integrity and reference integrity. We are continually adding new signals as behaviors change.

These signals are identified at all stages of the publication process, from submission to post-publication, allowing us to detect issues earlier and more comprehensively. There are also always humans in the loop, ensuring all papers flagged by our investigative technologies are reviewed by our in-house experts.

In our post-publication investigations, for example, we can better identify citation manipulation at the author, reviewer and editor levels; signs of organized misconduct by paper mills; AI-generated hallucinated references; conflicts of interest; duplicate submissions and more.

The speed of technology

Our newly developed technologies have notably improved our ability to manage large quantities of data, allowing us to assess millions of articles fast. For example, we can now evaluate more than 7,500 articles in an hour, and we are continually adding new signals to scan for. Using data science technology, we can quickly identify patterns indicative of unethical practices, supporting a more proactive approach to investigation. When substantial cases arise, robust processing power facilitates complete journal level audits spanning multiple years’ worth of content.

The speed of our new technologies helps us to solve cases faster while allowing more room for our experts to focus their insights. Authors are always given an opportunity to respond to allegations of misconduct, and our experts take extra care to make sure corrective action is taken only if necessary following a thorough investigation.

Expanding our expertise

As our tools have evolved, so has the scale of our expertise. While we have always included human oversight in our work, we’ve grown our Research Integrity & Publishing Ethics Center of Expertise since its founding in 2013 and broadened its expertise.

This team is comprised of trust and transparency technologists, engineers, data scientists, ethics experts, screening analysts and legal consultants, who work to prevent and identify ethical breaches, investigate potential research integrity cases, and help educate the publishing community about the impact of ethics issues. We share our knowledge and learnings with product development teams and others to make sure research integrity principles are integrated across all aspects of the business.

A broader range of tools

We are continuously improving and expanding the capabilities of our in-house tools and the variety of third-party tools we integrate into our systems.

We use both third-party and in-house tools for concerns such as image manipulation, plagiarism and artificially produced content. They help us detect large language model usage, whether required information or ethical approvals are missing, data availability, funding statements and more.

Nearly all Elsevier-owned journals are using a new in-house manuscript submission screening tool that enables our screening analysts to identify potential breaches of standards and policies that uphold integrity and ethics. Thus far, the tool has processed more than 1 million manuscripts, and of the manuscripts that surfaced a signal, over 20% were rejected.

For example, one of the signals identifies authorship changes at multiple points in the publication process. Our new Authorship Change Request Form provides extra safeguards for authors and editors and encourages transparency around changes in authorship.

Preventing fraudulent research and reducing retractions

While progress is leading to an increase in retractions as misconduct is identified in the published literature, these new improvements in technology and processes will likely reduce the need for as many retractions in the future. This trend reflects the tremendous advancements we are making in developing and employing sophisticated technology to combat fraud. Corrections to the scientific record will always occur, but perhaps one of the most exciting developments in the evolution of our technology is how we are moving beyond the detection of fraudulent research that has already been published to preventing fraudulent papers from making it through to publication.

We are optimistic that these improvements will help us identify and prevent misconduct earlier, enhancing the integrity of scientific publishing and strengthening confidence in research.