Comparing TR-Classifier and KNN by using Reduced Sizes of Vocabularies

International audienceThe aim of this study is topic identification byusing two methods, in this case, a new one that we haveproposed: TR-classifier which is based on computingtriggers, and the well-known k Nearest Neighbors.Performances are acceptable, particularly for TR-classifier,though we have used reduced sizes of vocabularies. For theTR-Classifier, each topic is represented by a vocabularywhich has been built using the corresponding trainingcorpus. Whereas, the kNN method uses a generalvocabulary, obtained by the concatenation of those used bythe TR-Classifier. For the evaluation task, six topics havebeen selected to be identified: Culture, religion, economy,local news, international news and sports. An Arabic corpushas been used to achieve experiments