Exécution de notre outil de synthèse sur un nouvel article
Vérifiez l’article ci-dessous, puis lancez sa synthèse.
An Optimized Preprocessing Algorithm That Enhances Efficiency and Effectiveness of Text Classification
H. Ramprasanth (Karpagam Academy of Higher Education), W. S. Vinisha (Karpagam Academy of Higher Education)
Résumé : In Natural Language Processing (NLP) the efficiency of the preprocessing methods have an important part in defining the sky with text classification. This study introduces advanced preprocessing algorithm which combine Orthogonal Matching Pursuit algorithm based on 3 different techniques for text classification. The basic idea behind this new type of preprocessing is usage of stopword removal and/or regular filtering in combination with tokenization and lowercasing. This is a holistic integration which not only provides a significant reducing in the feature dimensions of the data, but improves the overall quality of the matrix. Extensive simulation experiments on the well recognized 30 newsgroups data explicitly prove the advantages of the outlined method above and beyond the other cutting edge methods. In direct comparison, the new method, that was introduced in the current study, results in significant reductions of 25%, 35%, 26%, and 40% features, with obvious improvements in accuracy (20%, 25%, 35%, and 40%) on the four different data sets. Moreover, the introduced method covers exceptional achievements in text classification speed such as 22%, 35%, 33% and 43% gain for the corresponding datasets. To sum up the work, a synergy of stopword removal and the two other approaches in the preprocessing stage with the proposed Orthogonal Matching Pursuit algorithm can be seen as a powerful and holistic method for text classification. The empirical results give a good feeling to the effectiveness of the method and shows the deep reductions of the feature dimensionality, remarkable improvements on accuracy and a amazing speed-up on the classification speed which used to be associated with the use of current advancedapproaches.