Survey on Text Summarization

Abstract

Automatic text summarization is considered to be one of the hard problems because computationally there is no exact way of evaluating summary but the human can distinguish between good summary and bad summary. Also, summaries can be of various types like abstractive where new words and phrases are used, unlike extractive summarization where top scoring sentences from input text gets extracted as a summary sentence. Traditionally the focus of researcher was on building natural language generation which requires proper planning and realization of language. Various machine learning based approaches based on sequence labelling and SVR has been applied to extract summary sentences from the input text. Nowadays deep neural network models like sequence-to-sequence, LSTM, pointer-generator model are getting implemented to generate summaries. This report will give a brief idea about types of summary, summary evaluation measures and various ways to get summary.