TS-GAN with Policy Gradient for Text Summarization

Autor: Ashish Khanna, Viswanatha Reddy Allugunti, Nobel Dang
Rok vydání: 2021
Předmět:
Zdroj: Data Analytics and Management ISBN: 9789811583346
DOI: 10.1007/978-981-15-8335-3_64
Popis: Text summarization is a much evolving task, especially since neural networks were introduced. Similarly, generative adversarial networks (GANs) can be used to perform this task due to their ability to produce features or learn the whole sample distribution and produce correlated sample points. Thus, in paper, the authors exploited the characteristics of generative adversarial networks (GANs) for the abstractive text summarization task. The proposed generative adversarial model has three components: a generator which encodes the input sentences into much shorter representations; a discriminator which enforces generator to create understandable summaries; and a second discriminator which exerts upon generator to curb the output co-related to the input. The generator is optimized using policy gradient method, converting the problem into reinforcement learning. The ROUGE scores achieved by the model are as follows: R-1: 41.52, R-2: 16.20, R-L 37.21.
Databáze: OpenAIRE