New study maps algorithm influence via 40-year co-occurrence network
Deep learning extracts algorithm entities from full-text papers spanning 4 decades
A team of researchers from Yuzhuo Wang and collaborators has published a study titled "Exploring Academic Influence of Algorithms by Co-occurrence Network Based on Full-text of Academic Papers" on arXiv (June 2026). They use deep learning models to extract algorithm entities from the full text of academic papers in natural language processing (NLP), constructing overall, cumulative, and annual co-occurrence networks. The analysis covers more than four decades of publications, revealing that algorithm networks exhibit typical complex network features—dense interconnections that grow over time. Classic, high-performing algorithms and those sitting at the intersection of different research periods tend to score highest on centrality measures, indicating balanced influence across the field.
When an algorithm's influence declines, the network analysis shows it first loses its core network position, followed by weakening associations with other algorithms. This is the first large-scale analysis of algorithm co-occurrence networks, offering a temporal and structural view of how algorithms gain and lose academic influence. The findings provide a foundation for future research linking algorithms, scholars, and tasks, and could help identify emerging algorithms that bridge different research eras.
- Deep learning extracted algorithm entities from full text of NLP papers to build co-occurrence networks
- Covered 40+ years of publications with cumulative and annual network analysis
- Influence decline begins with loss of core network position before weakening connections
Why It Matters
Provides a data-driven method to track algorithm influence over time, guiding research investments and trend spotting.