Text Summarization Approaches Under Transfer Learning and Domain Adaptation Settings—A Survey

Abstract

AbstractData explosion is becoming a well-known concern as a result of rapid improvements in Internet-related technologies. The ease with which people can share and access information over the Internet has resulted in an abundance of data on any topic. Text summarization powered by deep neural networks can analyse vast volumes of text input and create a short summary on any topic to address the problem of information overload. However, deep neural networks require a lot of labelled data, and therefore, training them using tiny labelled text datasets is inefficient. Transfer learning and domain adaptation are possible solutions to this problem, and researchers have recently begun to investigate them in text domains. In this study, we analyse a number of recent articles that use transfer learning and domain adaptation to tackle the text summarization problem. The methodology for text summarization in transfer learning and domain adaption settings is explained in this survey. The application of transfer learning in sequence-to-sequence- (Seq2Seq) and non-Seq2Seq-based techniques for text summarization is discussed, with emphasis on crucial unique methodology and contributions from different works. Latest approaches for text summarization under different configurations of domain adaptation are discussed and compared. Issues, challenges and opportunities pertaining to this domain have also been identified and discussed for the benefit of the readers.KeywordsText summarizationTransfer learningDomain adaptationPretrained models