Automatic Generation of Web Article Headings Text with Heading Hierarchy

Abstract

When writing commercial web articles, one of the most common methods is to create a skeleton first, which is a set of headings like a table of contents, for the overall structure of the article, and then to write the text, figures, and tables according to this skeleton. Because such processes are manually performed, it is desired to automate the process using natural language processing technology to reduce the author's workload and achieve consistent quality. Many general sentence generation models take part of a sentence, especially the beginning of a sentence, and generate the rest of the sentence, which makes it difficult to use such models in the process of creating an article skeleton. We propose a sentence generation model that can be used to generate the skeleton of an article. By learning the hierarchy of article headings when building the sentence generation model, we were able to obtain as output a sequence of multiple headings with parent-child relationships that correspond to the skeleton of the article.