Abstractive Text Summarization Using Generative Adversarial Network and Relation Extraction

Abstract

In this paper, we propose a new adversarial model to solve some common problems existing in Generative Adversarial Network for text summarization. we simultaneously train a generative model G, a discriminative model D, and a model which auxiliary generator to generate a high quality summary .This module is like a teacher to guide G to quickly optimize himself and then against D, so we call it Teacher model. Teacher is an entity relationship extraction model to extract the triples of the real summary. The words in the triples are defined as keywords, and the keywords are revealed to the generator like the teacher points out the key points, and the summary generated by the guiding generator contains more Keywords. Model achieves competitive ROUGE scores with the baseline on CNN/Daily Mail dataset.