Deep Learning based Automatic Hindi Text Summarization

Abstract

Automatic text summarization (ATS) has become very popular among researchers because it involves generating a smaller subset of text from the main text. This subset represents the text’s entire and main idea. Natural Language Processing and Machine Learning are critical applications of ATS. Summarizations are classified into two types based on how they are generated: extractive and abstractive. In this project, we have implemented the Hindi text summarization, on which very little work has been done to date. It is a field where work can be proven helpful to the community if it is carried out. A machine learning model has been trained and tested through around 100,000 data entries so that it gives out very accurate summaries and, in turn, is very useful to society. The F-Score of the model is 58%, and the Rouge Score is 67.5%. Hence, we can consider that our model is decently accurate. Different libraries, like Pandas, NumPy, sklearn, etc., are used to develop the source code. The data model is trained using LSTM, word embedding, and seq2seq.