Question-based text summarization

Mengwen Liu

doi:10.17918/etd-7657

In the modern information age, finding the right information at the right time is an art (and a science). However, the abundance of information makes it difficult for people to digest it and make informed choices. In this thesis, we aim to help people who want to quickly capture the main idea of a piece of information before they read the details through text summarization. In contrast with existing works, which mainly utilize declarative sentences to summarize a text document, we aim to use a few questions as a summary. In this way, people would know what questions a given text document can address and thus they may further read it if they have similar questions in mind. A question-based summary needs to satisfy three goals, relevancy, answerability, and diversity. Relevancy measures whether a few questions can cover the main points that discussed in a text document; answerability measures whether answers to the questions are included in the text document; and diversity measures whether there is redundant information carried by the questions. To achieve the three goals, we design a two-stage approach which consists of question selection and question diversification. The question selection component aims to find a set of candidate questions that are relevant to a text document, which in turn can be treated as answers to the questions. Specifically, we explore two lines of approaches that have been developed for traditional text summarization tasks, extractive approaches and abstractive approaches to achieve the goals of relevancy and answerability, respectively. The question diversification component is designed to re-rank the questions with the goal of rewarding diversity in the final question-based summary. Evaluation on product review summarization tasks for two product categories shows that the proposed approach is effective for discovering meaningful questions that are representative for individual reviews. This thesis opens up a new direction in the intersection of information retrieval and natural language processing. Despite the evaluation on the product review domain, the thesis provides a general solution for question selection for many interesting applications and discusses the possibility of extending the problem to other domain-specific question-based text summarization tasks.

Question-based text summarization

Files and links (1)

Abstract

Metrics

Details

Question-based text summarization

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media