Finding a fitting and interesting title for a scientific paper can be difficult. While the tasksof paper-to-abstract and paper-to-review generation are active fields of research, the problem of generating a title from an abstract is - to our current knowledge - not tackled yet. In this paper, we propose a method to generate a set of titles, given an abstract, with the aim to inspire the author with different ideas for well readable titles. While the idea is to generate overall “good” titles, our focus lies on presenting different styles and not necessarily correct titles. Our models are based on OpenAI’s GPT-2 architecture which shows, among others, great capabilities in summarization tasks.
Data crawling is implemented in data crawling 1 and data crawling 2.