A New Supervised Term Weight Measure Based Approach for Text Classification

Abstract

The textual information is abundantly increasing in the internet through different types of social media platforms. Knowing the type of information is one challenging task to different information retrieval systems and researchers. Text classification is one research domain used to categorize the textual information into different classes. Most of the researchers proposed approaches based on the content used in the textual documents. Identification of appropriate terms for differentiating the text is one important task in text classification. After identification of terms for experiment, next very important task is determining the importance of a term in document representation. The term weight measures are used for finding the importance of a term in a document. In this work, a new supervised term weight measure named as TF-NRF-IPNDF-PNDDF is proposed. The performance of proposed term weight measure is compared with eight popular term weight measures such as TFIDF, TFIEF, TFRF, TF-IDF-ICSDF, TF-PROB, TF-IGM, CDallc and CDc. The experiment conducted on six standard classification datasets such as IMDB, HSS, FN, 20NG, AGN and CBN. Six different classification algorithms such as KNN, NB, LR, SVM, DT and RF are used for evaluating the performance of the proposed term weight measure. The proposed term weight measure attained best accuracies for different standard datasets compared with other term weight measures.