Hindi News Article's Headline Generation based on Abstractive Text Summarization

Text summarization is a technique that converts large text into shorter ones by keeping its important information. Most of the available summarizers, summarize English language text. The availability of summarizers in other languages is rare. In the proposed work, abstractive text summarization (ATS) for the Hindi language has been used to generate headlines from news text. To build the model, a Hindi news dataset has been used that contains different news articles related to sports, politics, crime, etc. After pre-processing the text, Word2Vec method has been used for word embedding. ATS model has been built using the Seq2Seq model with attention. After that, we have calculated evaluation metrics like ROUGE and BLEU scores of the result summaries. We have also fine-tuned two pre-trained models, indicBART and multi-lingual variant of T5. After result comparison, it was found that the indicBART perform better than the other two models, considered for the experiment.