Automatic Text Summarization: Methods, Metrics and Datasets

In this new age of big data, there is a variety of text data available of various lengths, which is unstructured in nature. The analysis of such unstructured data requires the use of Natural language processing (NLP) methods. Automatic text summarization (ATS) is an NLP tool leveraged to create short summaries of a longer text document or multiple documents. In this context, it is important for such models to capture all the significant factual information in the text to produce an efficient summary. There are mainly two types of summarization extractive and abstractive summarization. Extractive summarization ranks each sentence in the document based on its significance and a few important sentences form the final summary, however in abstractive summarization, based on the facts captured from the text, a set of new and concise sentences that includes all necessary facts, form the final summary. This work aims to discuss the recent trends in extractive and abstractive summarization methods. A set of evaluation metrics used for measuring the performance of ATS methods, along with some popular datasets used in training ATS models are also discussed briefly in this work.