Advertisement
Monday, November 16, 2026NewslettersAdvertiseWatch FL1
GET LOCALGo ad-freeHigh school sportsLocal eventsObituaries
News

Cornell researchers say today’s clicks can forecast tomorrow’s breakthroughs

Early downloads, likes and other online activity can help predict which research papers and software projects will remain influential years later, according to a Cornell study. The researchers developed forecasting models that used as little as one month of engagement data to estimate popularity five years into the future.

Cornell researchers say today’s clicks can forecast tomorrow’s breakthroughs

Early downloads, likes and other online activity can help predict which research papers and software projects will remain influential years later, according to a Cornell study.

The researchers developed forecasting models that used as little as one month of engagement data to estimate popularity five years into the future.

Finger Lakes Partners (Billboard)

The work applies a method called lead-lag forecasting to large datasets from arXiv, an online research repository, and GitHub, a platform used to develop and share software. Sarah Dean, an assistant professor at Cornell Bowers Computing and Information Science, was the senior author.

Yangfanyu Yang was scheduled to present the findings at the ACM Knowledge Discovery and Data Mining conference held Aug. 9-13 in Jeju, South Korea.

Millions of projects analyzed

The team analyzed anonymized download activity for 2.3 million papers on arXiv. Early download patterns were associated with the number of citations a paper received five years later, and established machine-learning models could make long-range forecasts using a short window of initial activity.

Researchers also studied about 938,000 GitHub repositories. Early pushes, which record additions or changes to a project, and stars, which users assign to projects they want to recognize or revisit, were correlated with the number of times a repository was copied through forks five years later.

The approach could help researchers and institutions identify promising work before conventional measures such as citations accumulate. The team said it may also offer a way to distinguish important papers from a growing volume of low-quality submissions, including material generated with artificial intelligence.

The bigger picture

Schools & Education

More reporting on this topic.

Explore this topic →

Earlier signals of emerging fields

Patterns showing which papers are downloaded together could reveal emerging fields or connections between areas of study. The researchers said similar methods could eventually be applied to social media, product sales and investing, where early attention may precede longer-term effects.

Kimia Kazemian and Zhenzhen Liu were co-first authors. Other contributors included Katie Luo, Sherman Gu, Moyun Du, Xinyu Yang, Jack Jansons, Kilian Weinberger and John Thickstun.

The work received support from the National Science Foundation, NASA, the PCCW Affinito-Stewart Award, a LinkedIn-Cornell research partnership, a Schmidt Sciences AI2050 fellowship and a collaboration with NewYork-Presbyterian.

Want to go ad-free?

Become an FL1 member and say goodbye to ads! Plus, you’ll be supporting a company rooted in the Finger Lakes for nearly 30 years.

$4.99 / month

Get ad-free access

Renews monthly. Cancel anytime through your account. Already contributing via Google? Sign in with your Google account and unlock ad-free access. Embedded YouTube videos and other third-party media players may still contain ads.

Start your day with FingerLakes1

Get the stories that matter from across the Finger Lakes, delivered to your inbox.

Your email will be added to our newsletter mailing list, hosted on Substack. Privacy policy