An Approach for Bengali News Headline Classification Using LSTM

Headline is called the soul of news. Headline carries a very important meaning. Generally, many people start reading the news after seeing the headlines. It is very important for the user to classify the headlines that he/she preferred. Classifying news type based on their headlines is a problem of text classification which lies under natural language processing (NLP) research. In different languages, there are many works done but none of them observed in Bengali. In our work we tried to visualize our very first approach to solve this problem. A LSTM-based architecture is used for solving this problem. A total of 4580 headlines are trained in our model. Finally, our model gives us 91.22% accuracy. The main challenge of our work is finding the right word vector. As far headlines are made up with different types of words and there is no similarity between any of them, so it is difficult to map them.