PaperBasket: Assessing Novelty and Integrity of Research

Whenever a research paper goes through the review process, the most difficult question to answer is whether the paper is novel. A reviewer does not have time to conduct an independent literature review and ends up trusting the authors who claim novelty for their work. PaperBasket automatically generates an independent literature review for each submission, providing the reviewer with a powerful tool to guide acceptance decisions.

Research

The research world is undergoing a dramatic transformation. Most research papers are now written with the help of Generative AI, ranging from subtle wording adjustments to complete fabrication of scientific results. Research quality assurance (i.e., paper reviewing), however, still relies on human labor. With the volume of submissions growing significantly across the board, reviewers are drowning in workload. Some resort to prompting GenAI to write the reviews. As a result, partially (or entirely) AI-generated papers are being automatically evaluated – a clearly unsustainable cycle. The research world is on the verge of collapsing in on itself.

We are building PaperBasket – a paper pre-screening system that assesses the novelty and integrity of paper submissions before they reach a human reviewer. Some of the questions our system will address include:

  • Is the submission similar to prior work published by others?
  • Is it an incremental follow-up to the authors’ own prior work?
  • Does it cite the most relevant prior work?
  • Is it situated in an overcrowded research area where significant contributions are unlikely?
  • Is it likely to be a genuine scientific breakthrough?
  • Is it likely to have been partially or fully fabricated?

A typical reviewer is capable of evaluating the quality of a submitted paper, but judging the paper’s novelty is a virtually impossible task given the dramatic increase in publication volume. Equipped with the PaperBasket report, the reviewer can make a data-driven, scientifically grounded decision about the review process for the submitted paper. If a very similar work has been published before, the submission might receive a desk reject on the grounds of plagiarism or lack of novelty. If the submission is flagged as potentially fabricated, the reviewer would not need to spend much time to detect obvious deficiencies. If, however, the paper is flagged as genuine and novel, it will definitely be worth the reviewer’s attention.

The first version of PaperBasket was developed in Fall 2025 by Syed Shahid, Oscar Callo von Platen, and Paul Ellameh as part of their Senior Capstone project under the supervision of Dr. Ron Bekkerman. On Dec 6, 2025, the PaperBasket team won the top recognition at the Capstone project competition!

Currently, PaperBasket compares an input paper with 2.4 million arXiv preprints. The efficiency of this comparison is our technological contribution. Once the most relevant preprints are identified for the input paper, Claude 4 analyzes the input paper in the context of those preprints. Similarities and differences between the input paper and its most relevant preprints are highlighted in the final report. The likelihood of the input paper having been automatically generated is also provided in the report. In the near future, we plan to significantly expand our database of published and preprinted research works to eventually achieve comprehensive coverage in our reports.

PaperBasket’s underlying dataset is an exciting playground for scientific research. With over 30 years of research papers from a variety of scientific disciplines, the dataset holds the secrets of the world’s technological innovation and scientific discovery. Those secrets are ready to be revealed. If you are interested in collaborating with us on PaperBasket, please reach out to Dr. Ron Bekkerman at ron.bekkerman@njit.edu.