Unsupervised Joint Learning for Headline Generation and Discourse Structure of Reviews

Abstract

This is an extension from a selected paper from JSAI2019. Recently, using a large number of reference summaries, supervised neural summarization models have achieved success. However, such data is rare, and trained models cannot be shared across domains. As a solution for such a problem, we propose the first unsupervised end-to-end headline generation model for a single review. We assume that a review can be described as a discourse tree in which the headline is the root and the child sentences elaborate on their parent. By estimating the parent from their children recursively, our model induces the tree and generates the headline that describes the entire review. Through the evaluation of the generated headline on actual reviews, our model achieved competitive performance with supervised models, especially on relatively long reviews. In induced trees, we confirmed that the child sentences explain the parent in detail and the generated headlines abstract for the entire review.