Horizontal and Longitudinal Comparisons Among AI Subfields: A Bibliometric Perspective

2026-05-09

AI Summary

This study addresses the lack of systematic comparative analysis on the long-term evolution of artificial intelligence subfields. Drawing on bibliometric methods, the authors construct a three-dimensional analytical framework encompassing research impact, collaboration networks, and author characteristics to conduct a structured comparison of 106,622 papers across five AI subfields—Computer Vision (CV), Machine Learning (ML), Web & Information Retrieval (Web&IR), Artificial Intelligence (AI), and Natural Language Processing (NLP)—from 2000 to 2024. Integrating expert interviews, multidimensional indicator selection, and advanced visualizations such as violin plots, chord diagrams, and Sankey diagrams, the study reveals for the first time AI’s evolutionary trajectory from unified diffusion toward structural differentiation. While all subfields have entered a phase of intense knowledge diffusion, they exhibit marked divergence: CV is strongly task-oriented; ML shows concentrated international collaboration but declining industry engagement; Web&IR remains stably industry-driven; AI as a whole sustains steady growth; and NLP maintains consistent development.

Abstract

Recent artificial intelligence has developed rapidly with significant interdisciplinary expansion, yet existing studies often treat it as a whole, lacking systematic long-term subfield comparisons and structural analyses, thereby limiting understanding of internal differences and evolutionary mechanisms. To address this gap, we employ bibliometric methods, using expert interviews and indicator screening to construct an analytical framework. Twelve bibliometric indicators are selected across three dimensions: Impact and Dissemination, Collaboration Characteristics, and Author Characteristics. We conduct horizontal and longitudinal analyses of five subfields (AI, CV, ML, NLP, Web\&IR) from 2000 to 2024. Using CSRankings classification and a dataset of 106,622 papers, we apply violin plots, chord diagrams, and sankey diagrams to characterize structural features and evolutionary paths. Results show that these subfields have entered high-intensity knowledge diffusion: academic impact increased, knowledge dissemination accelerated, external disciplinary reliance grown, and knowledge production shifted from closed accumulation to open, interdisciplinary, multi-actor networks. On this basis, subfields exhibit significant structural differentiation: CV leads in academic impact with a task-oriented trajectory; ML shows shrinking industry collaboration but concentrated international collaboration with a relatively dispersed structure; Web\&IR is strongly industry-driven with a stable collaboration network; AI shows continuous growth; NLP remains relatively stable. Overall, this study reveals artificial intelligence evolving from unified diffusion to structural differentiation, constructs an extensible multidimensional framework, and provides a quantitative approach for understanding complex technological field evolution.

Problem

• artificial intelligence
• subfields
• bibliometric analysis
• structural differentiation
• evolutionary mechanisms

Innovation

• bibliometric analysis
• structural differentiation
• interdisciplinary evolution
• longitudinal comparison
• AI subfields