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

A new study maps the collective influence of algorithms in natural language processing by constructing large-scale co-occurrence networks from more than four decades of full-text academic papers, revealing that the field’s algorithmic backbone has grown markedly denser over the past two decades [1][2]. The research, posted to arXiv on 23 June 2026, extracts algorithm entities using deep learning models and builds overall, cumulative, and annual co-occurrence networks to assess group influence through multiple centrality measures [1][2]. The authors report that algorithm networks display typical features of complex networks, with increasingly dense connections developing over approximately two decades [1][2]. Classic, high-performing algorithms and those situated at the intersections of different research periods tend to exhibit high popularity, control, centrality, and balanced influence [1][2]. When an algorithm’s influence declines, it usually loses its core network position first, followed by weaker associations with other algorithms [1][2]. The work is the first large-scale analysis of algorithm co-occurrence networks, covering more than four decades of academic publications [1][2]. It provides a temporal and structural view of algorithm influence and offers a foundation for future research on networks linking algorithms, scholars, and tasks [1][2]. The approach draws on bibliometric methods that have evolved since the 1960s, when Eugene Garfield’s Science Citation Index and Derek John de Solla Price’s citation network analysis established the basis for structured research on scientific metrics [5]. Citation analysis, a core bibliometric technique, constructs a network or graph representation of the citations shared by documents, a concept that now underpins tools such as Google’s PageRank algorithm [5]. The study’s network perspective also parallels social network analysis, which characterizes structures in terms of nodes and the ties that connect them and has been applied across disciplines from sociology to computer science [3]. By treating algorithms as nodes and their co-occurrence as edges, the authors adapt this framework to map the intellectual architecture of NLP research [1][2][3]. Concerns about algorithmic influence extend beyond academic impact. Algorithmic bias, defined as systematic and repeatable harmful tendencies in sociotechnical systems, can emerge from biased design, data selection, or unanticipated use, and has been documented in search engines, criminal justice, healthcare, and hiring [4]. The European Union’s Artificial Intelligence Act, adopted in 2024, represents one of the first legal frameworks to address such risks [4]. The new co-occurrence network study does not examine bias directly, but its structural mapping could help researchers trace how certain algorithms become entrenched and whether network position correlates with downstream effects [1][2][4].