Abstractive summarization incorporating graph knowledge

Automatic text summarization is an important challenge in natural language understanding. Automatic text summarization mainly includes extractive text summarization and abstractive text summarization. Extractive text summarization selects salient content from a document to form a summary, whereas abstractive summaries are formed by generating words and sentences. In this paper, we propose a novel abstractive summarization method incorporating graph knowledge. First, we propose a document word representation model based on a graph convolutional neural network for generating a summary. Then, the graph knowledge is integrated into an abstractive summarization model, which thus gains a better ability to generate new words. Finally, the abstractive summarization model is combined with a pointer generation model to solve the out-of-vocabulary problem. We apply our model to the Xsum and Gigaword summarization datasets, and the experimental results demonstrate that our model achieves state-of-the-art results on the Xsum dataset and results comparable to those of existing methods on the Gigaword dataset.