HSFO: Hunter Sail Fish Optimizer enabled deep learning for single document abstractive summarization based on semantic role labelling for Telugu text

Abstract

Abstract Accessing information from online has become difficult problem due to the internet's explosive rise in textual resources. User’s frequently sought for topic summaries from multiple sources to satisfy their informational demands. Single document abstractive summarization is significant role in Natural Language Processing (NLP) aiming to generate concise summary of source text from single document. Method of producing natural language summaries from text, keeping key points of the input document as such is a challenging task. In this research, hybrid optimization algorithm, namely proposed Hunter Sail Fish Optimizer (HSFO) is used leading to abstractive summarization. Here, acquired document is allowed for Semantic Role Labeling (SRL), at which Stanza tool is used for extraction of Predicate Argument Structures (PAS). Next to SRL, optimized features are generated by computation of semantic similarity using Wave-Hedges metrics. Moreover, semantic feature clustering of PAS is performed using Bayesian Fuzzy Clustering (BFC). Then, feature score is generated by utilization of HSFO for parameter selection, leading to abstractive summarization by Long Short-Term Memory-Convolutional Neural Network (LSTM-CNN). Here, HSFO is devised by combining Hunter-Prey Optimizer (HPO) and Sail Fish Optimizer (SFO). The dataset used in this work is Telugu dataset, from which text document in the sentence form is acquired. Finally, performance of HSFO_LSTM-CNN is analyzed by using four performance measures, precision, recall, F-measure, and Rouge, which shows superior values of precision as 0.887, recall as 0.925, F-measure as 0.906, and rouge as 0.815.