Metaheuristic Aided Improved LSTM for Multi-document Summarization: A Hybrid Optimization Model

Abstract

Multi-document summarization (MDS) is an automated process designed to extract information from various texts that have been written regarding the same subject. Here, we present a generic, extractive, MDS approach that employs steps like preprocessing, feature extraction, score generation, and summarization. The input text goes preprocessing steps such as lemmatization, stemming, and tokenization in the first stage. After preprocessing, features are extracted, including improved semantic similarity-based features, term frequency-inverse document frequency (TF-IDF-based features), and thematic-based features. Finally, an improved LSTM model will be proposed to summarize the document based on the scores considered under the objectives such as content coverage and redundancy reduction. The Blue Monkey Integrated Coot Optimization (BMICO) algorithm is proposed in this paper for fine-tuning the optimal weight of the LSTM model that ensures precise summarization. Finally, the suggested BMICO's effectiveness is evaluated, and the outcome is successfully verified.