PaperBasket: a Multi-Layer Knowledge Graph Platform for Detecting Academic Misconduct - Project Purpose As generative artificial intelligence (AI) tools increasingly penetrate research workflows, the research community faces new instances of misconduct, such as hallucinated references, AI-produced plagiarism, false authorship, and citation rings. Modern tools focus heavily on detecting AI-generated text, which is neither reliable nor an accurate indicator of actual misconduct. We intend to build a fully operational, AI-driven system to identify early indicators of research misconduct at a global scale. At the Leir Research Institute of the New Jersey Institute of Technology, we have developed PaperBasket.org which is a Big Data system for assessing novelty and integrity of research papers. PaperBasket screens a paper before it reaches a human reviewer and compares its full content with a massive repository of over 4 million research papers written by 5.5 million researchers. While PaperBasket provides underlying data infrastructure for misconduct detection, a semantic layer and a cognitive layer need to be built on top of it, which is the purpose of this project. Activities to be Performed To successfully perform misconduct detection, the project team will execute the following core activities Construct the Semantic Layer: Build a comprehensive, heterogeneous data network (the Knowledge Graph) that connects research papers, their authors, author affiliations, and publication venues - both factually (e.g. one paper cites another, two researchers are coauthors, a paper was published in a journal) and semantically (two papers are on the same topic, two researchers work on similar problems etc). Develop Automated Graph Analytics Tools: Build a fully automatic "fleet-of-AI- agents" component capable of running multiple analytical threads simultaneously across the Knowledge Graph to spot hidden anomalies. Design a Human-AI Collaboration Framework: Create an interactive framework where human operators and AI agents work together to analyze detected anomalies, filter out non-malicious activities, and isolate genuine academic misconduct. Within this framework, the constructed system will be tested on real-world scenarios, and an automatic testbed will be built to avoid analyzing similar cases in the future. Recommend Policy: Disseminate project findings and establish clear policy recommendations for how organizations can deploy systems such as PaperBasket to detect research misconduct. Expected Deliverables and Outcomes The primary deliverables include a completed heterogeneous semantic layer with time- stamped links, an automated anomaly detection system, an interactive human-agent review framework, and a published set of policy recommendations. The expected outcome is the ability to map out the dynamics of the research ecosystem and uncover otherwise invisible misconduct early in the review process. By analyzing data anomalies at the global scale, our system will successfully flag citation manipulations, identify AI-written papers, expose fake authors, track down research process flaws, etc. Intended Project Beneficiaries: - Academic and Industrial Research Labs that aim to safeguard their standing as trusted creators of stellar innovation. - Scientific Journals and Conferences that need to flag fraudulent manuscripts automatically before they reach human reviewers. - Research Sponsors and Funders that care about protecting the reputation and public trust in institutions that conduct, fund, and disseminate scientific research. Advancing ORI Priorities The project strengthens scientific integrity by automatically exposing deep anomalies in the academic publication ecosystem, at scale. It minimizes the administrative burden of journal editors and conference program chairs who will no longer have to waste their time on manual cross-referencing. Finally, it upholds the gold standard of science by replacing black-box AI models with an auditable human-agent review framework that will affect future research integrity policies.
