A semi-supervised Learning Approach to find equivalent long-string Organization Names

Frédérique Bordignon (1) , Nicolas Turenne (2) , Yann Feugueur

1 ENPC - École des Ponts ParisTech
2 LISIS - Laboratoire Interdisciplinaire Sciences, Innovations, Sociétés

Background: A platform called Opalia has been built to propose free access to all publications about a laboratory for a given range of years. This platform makes indexing of a corpus of a scientific article of a given lab. But in the French research system, a lab includes researchers from different organizations in the same unit generally called. UMR. Authors can write their laboratory names differently. Aim: Sorting a set of labels that is noisy can be seen as a binary classification into positives and leave negatives strings. We propose to use a cascade processing with the help of tagging some positive strings to build a relevant space of features that helps classification into good labels.

Domaines: Sciences de l'information et de la communication; Linguistique; Traitement du texte et du document

Frédérique Bordignon, Nicolas Turenne, Yann Feugueur. A semi-supervised Learning Approach to find equivalent long-string Organization Names. Colloque- Forum PEPS EXIA, Oct 2016, Champs sur Marne, France. 2016. ⟨hal-02310298⟩