New techniques for Arabic document classification

摘要: Text classi cation (TC) concerns automatically assigning a class (category) label to text document, and has increasingly many applications, particularly in the domain of organizing, for browsing large document collections. It is typically achieved via machine learning, where model built on basis collection features. Feature selection critical this process, since there are several thousand potential features (distinct words or terms). In cation, feature aims improve computational e ciency accuracy by removing irrelevant redundant terms (features), while retaining (words) that contain su cient information help with task. This thesis proposes binary particle swarm optimization (BPSO) hybridized either K Nearest Neighbour (KNN) Support Vector Machines (SVM) Arabic tasks. Comparison between approaches done using selected conjunction SVM, Decision Trees (C4.5), Naive Bayes (NB), classify hold out test set. Using publically available datasets, results show BPSO/KNN BPSO/SVM techniques promising domain. The sets also analyzed consider di erences types tend choose. leads speculation concerning appropriate strategy, based relationship classes categorization task at hand. investigates use statistically extracted phrases length two as cation. comparison Bag Words representation, alone TC decreases ers signi cantly combining bag phrase representations may increase SVM er slightly.