Automatic Marathi Text Classification

Abstract

Multifold growth of internet users due to penetration of Information and Communication technology has resulted in huge soft content on the internet. Though most of it is available in English language, other languages including Indian languages are also catching up the race rapidly. Due to exponential growth in Internet users in India common man is also posting moderate size data on the web. Due to which e-content in Indian languages is growing in size. This high dimensionality of e-content is a curse for Information Retrieval. Hence automatic text classification and structuring of this e-content has become the need of the day. Automatic text classification is the process of assigning a category or categories to a new test document from one or more predefined categories according to the contents of that document. Text classification works for 14 Indian languages are reported in the literature. Marathi language is one of the officially recognized languages of Indian union. Little work has been done for Marathi text classification. This paper investigates Marathi text classification using popular Machine Learning methods such as Naïve Bayes, K-Nearest Neighbor, Support Vector Machine, Centroid Based and Modified KNN (MKNN) on manually extracted newspaper data from sport’s domain. Our experimental results show that Naïve Bayes and Centroid Based give best performance with 99.166% Micro and Macro Average of F-score and Modified KNN gives lowest performance with 97.16% Micro Average of F-Score and 96.997% Macro Average of F-score. The proposed work will be helpful for proper organization of Marathi text document and many applications of Marathi Information Retrieval.