Evolutionary Algorithms and Neural Network-Based Fitness Functions for Extractive Text Summarization: A Comparative Study with ChatGPT

Anonymous

Abstract: Extractive text summarization deals with extracting a limited number of important sentences from a large document to create a summary. One novel approach already proposed in the literature is to model extractive summarization as an optimization problem, where a Genetic algorithm (GA) has been used for optimizing the selection of sentences from a text to generate the best extractive summary, which has been found outperforming state-of-the-art techniques. In this work, we build a similar model where apart from GA we used several different evolutionary algorithms (EA) in order to identify the combination that produces the best result. For this work, we have used different evolutionary algorithms, namely Discrete Differential Evolution (DDE), Cuckoo Search, Particle Swarm Optimization, and Firefly Search along with Genetic Algorithm, and have made comparison of their results with state-of-the art LLM viz. ChatGPT. The results are evaluated on the BBC news dataset using the precision-recall technique metric.

Paper Type: long

Research Area: Summarization

Contribution Types: NLP engineering experiment

Languages Studied: English