Algorithm Influence Mapped by Co-occurrence Networks in NLP.

This research delves into the academic influence of algorithms, particularly within the field of Natural Language Processing (NLP), by moving beyond isolated popularity metrics to analyze their collective impact. The study constructs extensive algorithm co-occurrence networks based on the full text of academic papers spanning over four decades. Utilizing deep learning models to extract algorithm entities, it builds overall, cumulative, and annual networks to observe their structural characteristics and evolution. By applying various centrality measures, the study assesses the group influence of algorithms across the field and over time. Findings indicate that these algorithm networks exhibit typical complex network features, with increasing density over two decades. Classic, high-performing algorithms, especially those at the intersection of different research periods, tend to demonstrate high popularity, control, and balanced influence. The research also reveals that an algorithm's decline in influence typically begins with a loss of its core network position, followed by weaker associations with other algorithms, providing a temporal and structural view of algorithmic impact.