On Improving Text Generation Via Integrating Text Coherence
Lisi Ai, Gao Baoli, Jianbing Zheng, Ming Gao · 2019
Automatic text generation techniques with either extractive-based or generative-based methods are becoming increasingly popular and widely used in industry. In contrast to existing extractive-based text generation approaches that ignore the text coherence, our proposed approach integrates both keyword coverage and text coherence into our optimization framework. In this paper, we employ the semantics-based coherence and syntax-based coherence metrics to evaluate the text coherence. Extensive experiments on a real corpus demonstrate that our method outperforms baselines overall regrading ROUGE and Human evaluation metrics. Our model provides new insights on how to utilize coherence measures to arrange the sentences extracted by keyword covering method. The proposed method has been deployed on a real system to help generate coherent text.