Authors: Didier Torny ¹; Laurent Capelli ^2,3; Lydie Danjean ^2,3; Stéphane Pouyllau ^2,3

• 1 Centre de Sociologie de l'Innovation i3
• 2 Huma-Num : la TGIR des humanités numériques
• 3 Huma-Num : l'IR* des humanités numériques

Although bibliometrics and library science are older, bibliometric tools were really born about 50 years ago and were only made available to a large audience with the widespread use of the Internet. Although their concrete forms have been largely modified, they are still based today on epistemic and computer foundations decided at the time. Three important characteristics of these tools can be identified: first, they are proprietary, i.e. users not only have to pay for access to the data but it is also difficult to manipulate and verify; second, in the name of a principle of scarcity or quality, tool creators assume to rely only on a selection of accessible scientific documents; thirdly, this choice of a small sample is, moreover, very marked by a historical irreversibility that makes invisible in particular some types of documents (books, conferences, preprints) and written documents in the vast majority of languages other than English. However, over the last twenty years, there has been a progressive liberation of scientific texts through the existence of different disciplinary (ArXiv, PubMedCentral, REPEC) and institutional (HAL, universities archive...) open archival systems, and publication models allowing the harvesting of texts and/or metadata - including the references cited. It is in the continuation of this movement that the construction of a real tool, Matilda, is taking into account all available sources and user personalization, in order to serve as an elementary brick for bibliographic and bibliometric research in the age of open science.