Towards an Automatic Generation of Persuasive Messages

摘要

In the last decades, the Natural Language Generation (NLG) methods have been improved to generate text automatically. However, based on the literature review, there are not works on generating text for persuading people. In this paper, we propose to use the SentiCAN framework to generate messages that are classified into levels of persuasiveness. And, we run an experiment using the Microtext dataset for the training phase. Our preliminary results show 0.78 of novelty on average, and 0.57 of diversity in the generated messages.