Abstractive Text Summarization for English Language Using NLP and Machine Learning Approaches

A text summarizing system’s principal goal is to extract the most important information from a given text. We are building an abstractive text summarizer (ATS) to generate high-quality English summaries that is similar to human generated summaries. We have worked on three approaches in our 1st approach, we have used a bi-directional LSTM and encoding layer (with attention mechanism) and attention model in decoding layer, and to generate a summary, we have applied the sequence-to-sequence model. We have used the concept of embedding layers, tokenization, attention mechanism, stacked BiLSTM, and Seq2Seq model. Our 2nd approach incorporates RoBERTa which is robustly optimized BERT pre-training. In our 3rd approach, we have used the DistilRoBERTa model. RoBERTa and DistilRoBERTa optimize the training of BERT architecture in order to take less time during pre-training. The ROUGE score was used to test our model, and the results were compared. The DistilRoBERTa model does better than other models for text summarization problem.