ForcesAtlas: construction of a multi-source, AI-assisted evidence discovery database of Armed Forces research

Finding the right information in a vast library of scientific papers is a task that grows harder every day. Researchers, doctors, and policy makers rely on this evidence to make decisions about health, safety, and support for people in uniform. However, the literature concerning military personnel, veterans, and their families is scattered across many different digital shelves. Some databases focus on medicine, others on psychology, and still others on social issues. Furthermore, the people being studied are described in many different ways—sometimes as "active duty," other times as "veterans," "reservists," or "military spouses." This inconsistency means that a simple search for "soldiers" might miss a crucial study about "former service members," while a search for "military" might accidentally pull up articles about military hardware or strategy that have nothing to do with human health. The challenge is not just finding the papers, but finding the ones that actually matter to the people who serve and those who have served.

To solve this problem, a team of researchers from universities in the United Kingdom and the United States built a new digital tool called ForcesAtlas. Their goal was to create a single, constantly updated collection that gathers relevant research from across the entire scientific world, regardless of the specific field or the words used to describe the people involved. Instead of forcing a user to build a complex list of search terms every time they want to investigate a topic, ForcesAtlas does the heavy lifting beforehand. It scans multiple major scientific databases, identifies articles that likely concern the armed forces community, and organizes them into a dedicated space. The system uses artificial intelligence to read the titles and summaries of millions of articles, deciding which ones are relevant based on their content rather than just matching keywords. This allows a researcher to ask a question about, for example, housing stability or mental health, and immediately see a curated list of studies that are actually about military-connected people, without having to filter out thousands of unrelated results.

The team constructed this database by pulling data from several large sources, including PubMed for medical literature, Europe PMC for life sciences, and other broad scholarly networks. They built a system that checks these sources weekly, looking for new articles. When a new paper is found, the system compares it against what it has already seen to ensure there are no duplicates. Then, a sophisticated computer program reads the article's title, publication year, and abstract to judge its relevance. The system assigns a rating of high, medium, low, or none to indicate how strongly the paper connects to the armed forces community. A high rating means the paper is central to the topic, while a medium rating suggests a direct connection within a broader study. To ensure the computer is not making mistakes, human reviewers check a sample of the decisions. In a test involving 2,000 articles, the human reviewers agreed with the computer's classification more than 92 percent of the time, showing that the system is highly reliable at distinguishing relevant research from the rest.

The result is a massive, searchable library that currently holds over 2.4 million article records, with about 86,000 of these publicly available for immediate use. The collection is heavily weighted toward research from the United States, reflecting where much of the available literature originates, but it also includes significant work from the United Kingdom, Afghanistan, Iraq, and other nations. The most common topics found in the database relate to veterans, mental health, and post-traumatic stress, though the scope covers everything from physical health and employment to family life and criminal justice. To prove that the database actually works, the researchers tested it against five recent, high-quality reviews that had already been published on specific military topics. They checked to see if the papers included in those reviews were also present in ForcesAtlas. The new database successfully found nearly 88 percent of the known important papers, with some specific tests showing it found every single relevant article. This suggests that the tool is effective at capturing the evidence that experts are already using, while also finding additional material that might otherwise be missed.

ForcesAtlas is designed to complement existing resources rather than replace them. It does not claim to be the final word on every subject, but it offers a powerful starting point for anyone trying to understand the experiences of military communities. By moving the difficult work of identifying relevant populations to the beginning of the process, the tool frees researchers to focus on the actual questions they want to answer. Whether a user is a student looking for a thesis topic, a clinician seeking the latest treatment guidelines, or a policy maker evaluating a new program, the database provides a clear path through the noise of the wider scientific world. It stands as a practical example of how modern technology can be used to organize knowledge, ensuring that the evidence needed to support those who serve is easier to find, understand, and use.