Abstractive Text Summarization for Legal Documents Using Optimization-Based Bi-LSTM with Encoder-Decoder Model

In recent years, the amount of textual data has grown quickly, creating a useful resource for information extraction and analysis. Due to the high complexity and unstructured nature of legal documents, automated text summarizing (ATS) is a necessary but difficult process. ATS is a technique that uses computer power to summarize lengthy paragraphs quickly. Summarizing the extensive materials manually is a highly difficult and time-consuming operation for people. As a result, an optimization-based deep learning (DL) model for abstract summarization is presented in the paper (AS). The proposed working procedure is broken down into three stages. They are text pre-processing, feature extraction, and abstractive summary categorization. The dataset is first placed through a pre-processing process, including stop word removal, tokenization, lemmatization, and stemming. The words and phrases are then represented in vector format throughout the feature extraction phase, which is handled by the NumberBatch (a combination of Enhanced GloVe Model (EGM), Enhanced FastText (EFT), and word2vector). To produce the abstractive text summary, the features obtained from the Numberbatch are fed into the DL model Bi-LSTM-based encoder-decoder with attention model. The metaheuristic optimization Honey Badger algorithm (HBA) is employed to optimize the network weights. This would increase the effectiveness of creating summaries based on ROUGE scores, and the proposed Bi-LSTM-HBA model performs better than currently used methods.