Reproduce “Dynamic technology impact analysis: A multi-task learning approach to patent citation prediction” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Youngjin Seol, Jaewoong Choi, Seunghyun Lee, Janghyeok Yoon

Researchers from Konkuk University and KIST introduced a Multi-Task Learning (MTL) framework to predict the dynamic impact of technologies by forecasting patent citations across short, mid, and long-term horizons. This system improved the identification of high-impact technologies, particularly for mid-term predictions, and provided insights into which patent characteristics drive technological influence at different stages.

Key insights

  • Technology impact is inherently dynamic, with patents demonstrating distinct temporal patterns such as 'sustained-impact,' 'peak-and-fade,' and 'late-blooming' over their lifecycles.
  • The MTL approach significantly outperforms Single-Task Learning models in identifying 'valuable' and 'breakthrough' technologies by effectively leveraging the interdependencies between different temporal impact predictions.
  • SHAP analysis reveals that the drivers of technology impact change over time; 'scientific knowledge' is critical for short-term, 'growth speed' for mid-term, and 'technological breadth' and 'international priority range' for long-term breakthroughs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.