version française rss feed
s'authentifier
Fiche détaillée  Récupérer au format
Congrès LREC 2000, "Second international conference on langage ressource and evaluation"-, Athènes, 29-2 juin 2000 : (2000)
Liste des fichiers attachés à ce document : 
PDF
sic_00000335.pdf(67.5 KB)
sic_00000335, version 1
Cardinal , nominal or ordinal similarity measures in comparative evaluation of information retrieval process
Christine Michel
(01/06/2000)

Similarity measures are used to quantify the resemblance of two sets. Simplest ones are calculated by ratios of the document's number of the compared sets. These measures are simple and usually employed in first steps of evaluation studies, they are called cardinal measures. Others measures compare sets upon the number of common documents they have. They are usually employed in quantitative information retrieval evaluations, some examples are Jaccard, Cosine, Recall or Precision. These measures are called nominal ones. There are more or less adapted in function of the richness of the information system's answer. Indeed, in the past, they were sufficient because answers given by systems were only composed by an unordered set of documents. But usual systems improve the quality or the visibility of there answers by using a relevant ranking or a clustering presentation of documents. In this case, similarity measures aren't adapted. In this paper we present some solu tions in the case of totally ordered and partially ordered answer. Similarity measures are used to quantify the resemblance of two sets. Simplest ones are calculated by ratios of the document's number of the compared sets. These measures are simple and usually employed in first steps of evaluation studies, they are called cardinal measures. Others measures compare sets upon the number of common documents they have. They are usually employed in quantitative information retrieval evaluations, some examples are Jaccard, Cosine, Recall or Precision. These measures are called nominal ones. There are more or less adapted in function of the richness of the information system's answer. Indeed, in the past, they were sufficient because answers given by systems were only composed by an unordered set of documents. But usual systems improve the quality or the visibility of there answers by using a relevant ranking or a clustering presentation of documents. In this case, similarity measures aren't adapted. In this paper we present some solu tions in the case of totally ordered and partially ordered answer.
Sciences de l'Homme et Société/Sciences de l'information et de la communication/Ingénierie des systèmes d'information

Sciences de l'Homme et Société/Sciences de l'information et de la communication/Communication et information scientifique
mesure de similarité – ordre – ordonnancement – recherche d'information – évaluation

tous les articles de la base du CCSd...