Transformer-based Hierarchical Topic-to-Essay Generation

With the development of 5G network and Internet of things (IOT), a large amount of information is required. In this work, we focus on Topic-to-Essay Generation (TEG), which aims to generate the text based on the topics. Existing methods utilize the RNN-based models, and it's not so useful to capture the long dependencies in the text. The information of topics is insufficient to generate a long text and the existing methods also suffer from the problem about the topic relevance of the text. To fill these gaps, we propose a Transformer-based Hierarchical Topic-to-Essay Generation Model (THTEG), and the experimental results on a real dataset show that our model performs better than the baselines in terms of automatic evaluation and human evaluation.