An Efficient Document Categorization Approach for Turkish Based Texts

Abstract

0

: Since, it is infeasible to classify all the documents with human effort due to the rapid and uncontrollable growth in textual data, automatic methods have been approached in order to organize the data. Therefore a support vector machine (SVM) classifier is used for text categorization in this study. In text categorization applications, the text representation process could take a huge computation time on weighting the huge size of terms. So far, lexicons that contain less number of terms are used for the solution in the literature. However it has been observed that these kinds of solutions reduce the accuracy of the text classification. In this paper, the term-document matrix is constructed as user dependent according to the purpose of classification. Since the number of terms is still relatively large, we used a hash table for efficient search of terms. Hereby an efficient and rapid TF-IDF method is introduced to construct a weight-matrix to represent the term-document relations and a study concerning classification of the documents in Turkish based news and Turkish columnists is conducted. With the proposed study, the computational time that is required for term-weighting process is reduced substantially; also 99% accuracy is achieved in determination of the news categories and 98% accuracy is achieved in detection of the columnists.