Time-aligned graphs improve predicting scientific relations over time
Time-Aligned Evolving Concept Graphs for Scientific Relation Forecasting
Artificial IntelligenceInformation Retrieval
Summary
Predicting which scientific ideas or concepts will connect in the future can help guide new discoveries. The authors present a method that models how both the meanings of concepts and their connections change together over time by using the publication dates of papers as update points. This approach aligns the evolving meanings and relationships, improving the accuracy of forecasting when and how scientific concepts will relate. Their method showed better results on a large dataset of scientific papers compared to earlier techniques.
What this means in practice
- •For research data engineers: Improve tools that track and predict emerging connections in scientific literature to guide research discovery processes.
- •For knowledge graph developers: Enhance dynamic knowledge graphs by jointly updating concept meanings and relationships aligned with publication timelines.
Authors
Fred Sun, Jingze Wang, Minkun Xu, Shangqi Guo
Abstract
Forecasting scientific relations can guide discovery by identifying promising connections before they emerge. Existing approaches often model concept semantics and graph structure separately or summarize semantics over coarse historical snapshots, leaving semantic representations potentially misaligned with rapidly evolving graph evidence. We propose a time-aligned evolving concept graph framework that jointly models semantic and structural evolution. Its core idea is to treat dated papers as shared update events, reconstructing semantic and structural states from the same publication history through each prediction time. Pair-level fusion combines these states to forecast first co-occurrence, relation formation, and conditional relation type. Holding architecture and training fixed, refreshing context alongside graph updates improves mean relation AUPRC by 16.6% over frozen context. On a graph built from 187,848 papers with 270,687 concepts and 7.45 million co-occurrence links, the complete framework improves mean relation AUROC from 0.9290 for the strongest evaluated baseline to 0.9722, with mean population-weighted AUPRC 0.005778.