Recent decades multi-level Bangla text classification is very crucial task for Bangla newspaper portal to improve their recommendation system as well as manual labour to categorize their different types of document. In the field of natural language processing and text mining a very few work has been done due its limited resource. The goal of this paper to provide a standard solution to overcome this constraints. In this paper a large number of dataset which consists almost 400k Bangla newspaper articles as JSON data format have utilized. We are using supervised machine learning model which consists of Long Short Term Memory (LSTM) for data extraction and data cleaning preprocessing method for Convolutional Neural Network (CNN). These newspaper articles have been categorized into nine categories named Bangladesh, Opinion, Internatioal, Education, Economy, Technol-ogy, Sports, Life-Style and Entertainment. Finally we understand the findings obtained from the model presented by various researcher’s and prove that our model is more reliable than the previous framework.