Extracting Crucial Information from Lengthy Transcripts Using Text Summarisation

The advancement of technologies pays the way for plenty of information to be from online. Researchers are presently inspecting advanced techniques that develop deep learning models, comprising Bidirectional Encoder Representations from Transformers (BERT), to overcome this problem. However, because of the limitations caused by the length of the document, BERT is not able to amply condense large documents. This research suggests a novel procedure to address the complications by adding a grouping of methods of pre-processing technique, Feature Extraction (FE), and classification. In FE, the focus is to extract the important features using a Cluster-Based System (CBS) and balance the dataset and word representation. Finally, the BERT is pre-trained on enormous datasets and is then united with the Long Short-Term Memory (LSTM). Experimental exploration is carried out using datasets from arXiv and YouTube revealing that the proposed method implemented better than several representations. The arrangement of LSTM with BERT has been given outcomes in a noteworthy enhancement in the text summarisation. In addition to being favourable to a variety of fields that demand summaries that are precise and informative, this research establishes opportunities for summarising long transcripts to give valid points from the keywords.