Development of a Bangla news classification system

Abstract

On-line newspapers and digital editions of print newspapers has become more and more popular as technology continues to grow. As the number of popular news articles grows and people also have different interests they want to categorize news to read only their interested topics. Classification of on-line news in the past, has often been done manually. Text classification is a well-studied problem. Several methods have been proposed and many of them can be directly applied to news classification as long as there exists a good set of training documents for each predefined category. To develop a news classifier we can use either Collaborative filtering, Content-based filtering, Subscription-based personalization approach. Among the above three approaches, we have chosen to adopt Content filtering approach to support personalized news classification in Categorizer system. The main difficulty in using this approach is that the ve -Gram Based Text - Bangla text classification. We want to Compare different Classifier algorithm to know better classification result and finally build a web application where users can search and read category based Bangla news articles. In this project an application has been developed to compare different classification algorithms for Bangla news classifier and a web application for categories news. Data collection, analysis and model building part has been developed with Python and Python based machine learning library (Scikit Learn). For development the application I have used Python based web framework. After analysis different algorithms we found Naive Bayes has more accuracy then other machine learning algorithms.