A Survey on Transformer-Based Extractive Summarization Methods

Searching for a topic on the internet exposes the user to a large amount of textual data. Reviewing all these documents is not feasible in terms of time. Text summarization is one of the active areas in the field of natural language processing. This area is divided into two main categories: extractive and abstractive. Extractive summarization identifies informative and salient sentences from the text and outputs them as the result. Abstractive summarization paraphrases the text and generates a summary with new sentences. Various methods have been proposed for both types so far, including statistical methods, machine learning-based methods, deep learning, and more recently, transformer models. The transformer model is a neural network-based architecture that has achieved promising results in various tasks in natural language processing, including extractive summarization. This paper provides a comprehensive survey of transformer-based extractive summarization models for researchers and describes the evaluation measures.