Presenting an improved combination for classification of Persian texts

Since text mining saves a large amount of information in text format, it has a very high potential application. One of the main applications of text mining is to classify texts in subject order. In this paper, we tried to propose a new method in order to increase classification accuracy and efficiency, by considering different methods of Persian text classification. We used a number of 5330 news of Hamshahri data collection, for classification. In pre-processing of texts for removing stop words, we proposed a new method by using entropy of words. To extract the feature, word frequencies, and Tf-idf methods have been used. K nearest neighbor algorithm, Naïve Bayes classification, and mixture of classifiers , have been used to classify texts, by using combinational classification and mixture of experts. Implementation of proposed method has caused a 15 percent improvement comparing to the previous works done on this data collection, by presenting entropy in pre-processing and also mixture of classifiers. In the best condition, scientific and cultural news has gained 96.36 percent classification accuracy.

Keywords:

Text mining , classification , entropy , mixture of classifiers , Naïve Bayes classification , k nearest neighbor , Tf-idf