La détection de communautés est une question centrale en analyse de réseaux. Cet article combine une approche socio-historique à la reconstruction expérimentale de programmes informatiques afin d'éclairer l'histoire des premiers algorithmes de détection de cliques, problème qui compte encore aujourd'hui parmi les problèmes NP-complets non résolus. Restituer les recherches menées par l'archéologue Jean-Claude Gardin depuis les années 1950 sur le traitement de l'information non numérique et l'analyse de graphes met en évidence ces contributions précoces à l'informatique réalisées depuis les sciences humaines et sociales. Ces applications originales de l'informatique aux humanités ont reçu une réception et une reconnaissance limitées. Ce fait est éclairé par deux facteurs : 1) les politiques de financement, qui ont motivé le transfert des efforts de recherche sur les graphes depuis un éphémère espace interdisciplinaire vers des organisations de recherche en informatique, domaine alors émergent ; 2) les carrières erratiques des algorithmes, où l'efficacité, les erreurs, les corrections et le statut des auteurs ont été des facteurs déterminants. Ces facteurs se combinent aux effets des historiographies et des bibliographies sur la conservation, la découvrabilité et la réutilisation des résultats scientifiques.
From as early as the 1950s, J.C. Gardin's work spanned both archaeology and the emerging automation of numerical computation and documentation. In 1961, with P. Garelli, he published the first automated application of graph theory to historical materials, working from Assyrian cuneiform tablets documenting economic relations. This work was then widely ignored both in archeology and network analysis. However, in the past twenty years, socio-epistemic claims related to the growth of the Internet and computing (digital humanities, computational archaeology, etc.) have brought a surge of interest in Gardin's work, which is now regarded as pioneering. Working from archive materials and publications, this paper shows how a historical sociology of scientific writings can be relevant to the history of automation in historical sciences. The paper examines Gardin's recognition as an influential forerunner of computational archeology, showing that : 1) although Gardin had access to resources (financial, instrumental, etc.) that were rare at the time, and could have provided material for the foundation of a school or a specialty, he did not however pursue this ambition; 2) the demonstrative purposes pursued by Gardin with his study of 1961 economic networks varied between the 1960s (demonstrating the relevance of non-numerical computation) and the 1980s (legitimizing simulation in the social sciences), but were never concerned with network analysis as such.
Three geographers who have defended their thesis in 2021 and who use network analysis methods from other disciplines (sociology, computer science, physics) explain the choice of methods, indicators and software used in their work. They insist on both the weakness of initial training and the crucial role of self-training. The strengths and weaknesses of network analysis and link-node visualizations are finally discussed.