Exploring Academic Influence of Algorithms by Co-occurrence Network Based on Full-text of Academic Papers

2026-06-23 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

This study constructs and analyzes large-scale algorithm co-occurrence networks in Natural Language Processing (NLP) to explore the collective academic influence of algorithms. Utilizing deep learning models, researchers extracted algorithm entities from the full text of academic papers, building overall, cumulative, and annual networks. The analysis, covering more than four decades of publications, revealed that these algorithm networks exhibit typical complex network features, with connections becoming increasingly dense over approximately two decades. Findings indicate that classic, high-performing algorithms and those positioned at the intersections of different research periods demonstrate high popularity, control, centrality, and balanced influence. When an algorithm's influence wanes, it typically loses its core network position before its associations with other algorithms weaken. This research represents the first large-scale analysis of its kind, offering a temporal and structural perspective on algorithm influence.

Key takeaway

For research scientists evaluating algorithm longevity or impact, this analysis reveals that sustained influence depends on maintaining a core network position. You should consider network centrality metrics, not just citation counts, when assessing an algorithm's true academic footprint. Prioritize algorithms with robust connections across research periods. These are more likely to retain relevance and control within the field.

Key insights

Algorithm influence in NLP can be effectively analyzed by constructing and examining large-scale co-occurrence networks from academic paper full texts.

Principles

• Algorithm networks exhibit complex features.
• Influential algorithms hold central network positions.
• Declining algorithms first lose core network status.

Method

Deep learning models extract algorithm entities from full texts. These form overall, cumulative, and annual co-occurrence networks. Multiple centrality measures then assess group influence and structural characteristics.

Topics

• Algorithm Co-occurrence
• Network Analysis
• Natural Language Processing
• Deep Learning
• Academic Influence
• Centrality Measures

Best for: AI Scientist, Research Scientist, NLP Engineer

Related on AIssential

• Most Influential AISTATS Papers (2026-03 Version) — Paper Digest's AISTATS ranking highlights influential research based on citations from papers and patents, offering dynamic insights into AI/statistics trends.
• Most Influential ArXiv (Optimization and Control) Papers (2026-04 Version) — Influence in optimization and control research is dynamically measured by citations from both papers and patents.
• Merit or networks? What decides where research is published — Scientific publishing outcomes are determined by a complex interplay of merit (idea and execution quality) and author networks.
• From sunblock to softblock: Analyzing the correlates of neology in published writing and on social media — Neology correlates similarly across published writing and social media, though topic popularity's influence varies.
• Graphs of Research: Citation Evolution Graphs as Supervision for Research Idea Generation — Citation evolution graphs provide effective supervision for LLM-based research idea generation.
• My Eureka moments in research — Personal "Eureka moments" in NLG research often correlate with influential, highly-cited papers and enduring insights.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.