Abstractive text summarization using Transformer based deep learning

No ratings

Presented at NCS2019 2020 by

Xin Liu

Extractive and abstractive are two main text summarization techniques, unlike previous works which treat them as two separated tasks. In this paper,we present TRCC-ES, an original model based transformer with copy and coverage on extractive sentences. We aim to obtain a short summary with a precise text span in a long paragraph. On the one hand, we combine extractive model with abstractive model to generate a more readable paragraph by calculating word-level attention after obtaining sentences with high ROUGE scores. On the other hand, we apply transformer language model to generate summarization. The results of the experiments based on two abstractive summarization datasets show that our model significantly outperforms the state-of-the-art summarization models.