Multilingual documents clustering based on closed concepts mining

Abstract

The scarcity of bilingual and multilingual parallel corpora has prompted many researchers to accentuate the need for new methods to enhance the quality of comparable corpora. In this paper, we highlight the interest and usefulness of Formal Concept Analysis in multiligual document clustering to improve corpora comparability. We propose a statistical approach for clustering multiligual documents based on multilingual Closed Concepts Mining to partition the documents belonging to one or more collections, writing in more than one language, in a set of classes. Experimental evaluation was conducted on two collections and showed a significant improvement of comparability of the generated classes.