Savoy, JacquesBorel, AlainKrause, Jan2009-10-132009-10-132009-10-132009https://infoscience.epfl.ch/handle/20.500.14299/43674Due to the multiplication of digital bibliographic catalogues (open repositories, library and bookseller catalogues), information specialists are facing the challenge of mass-processing huge amounts of metadata for various purposes. Among the many possible applications, determining the similarity between records is an important issue. Such a similarity can be interesting from a bibliographic point of view (i.e., do the records describe the same document, the answer to which can be useful for deduplication or for collection overlap studies) as well as from a thematic point of view (suggestion of documents to the user, as well as content management within the framework of a library policy, automatic classification of documents, and so on). In order to fulfil such various needs, we propose a flexible, open-source, multiplatform software tool supporting the implementation of multiple strategies for record comparisons. In a second step, we study the relevance and performance of several algorithms applied to a selection of collections (size, origin, document types...).Information retrievalBibliographic similarity and duplicate detectionMARCXMLPythonstudent work::master thesis