EXPERIMENTS WITH ARABIC TOPIC DETECTION
Rim Koulali, Abdelouafi Meziane · 2013
The continuous growth of information on the Internet and the availability of a large mass of electronic documents in Arabic language make Natural Language processing (NLP) tasks play an important role to enhance and facilitate the access and the exploitation of information. Among available NLP tasks, we are interested in Arabic Topic Detection. Our objective is to realize an indexing system capable of identifying the general topics discussed in Arabic unvowelized documents. The proposed topic detection system of Arabic texts is based on Mutual Information for Topic Oriented Vocabulary (TOV) and classification according to Jaccard and adapted TF-IDF indicators. The experimental results are presented in terms of precision, recall and F1 measure evaluating the influence of factors such as: vocabulary length and morphological analysis on Arabic Topic Detection.