Summarizing Lengthy Questions

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 792–800, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP Summarizing Lengthy Questions Tatsuya Ishigaki Hiroya Takamura Manabu Okumura Tokyo Institute of Technology ishigaki@lr.pi.titech.ac.jp, {takamura, oku}@pi.titech.ac.jp Abstract In this research, we propose the task of question summarization. We first ana- lyzed question-summary pairs extracted from a Community Question Answering (CQA) site, and found that a proportion of questions cannot be summarized by ex- tractive approaches but requires abstrac- tive approaches. We created a dataset by regarding the question-title pairs posted on the CQA site as question-summary pairs. By using the data, we trained extrac- tive and abstractive summarization mod- els, and compared them based on ROUGE scores and manual evaluations. Our exper- imental results show an abstractive method using an encoder-decoder model with a copying mechanism achieves better scores for both ROUGE-2 F-measure and the evaluations by human judges. 1 Introduction Questions are asked in many situations, such as conference sessions and email communications. However, questions can sometimes be lengthy and hard to understand, because they often contain pe- ripheral information in addition to the main focus of the question. To address this issue, we propose the task of question summarization; summarizing a lengthy question into a simple question that con- cisely represents the original content. As an example of an excerpt from a Commu- nity Question Answering (CQA) site, Yahoo An- swers1, is shown in Table 1. In this example, the gist of the question is whether the chlorine will stripe the questioner’s hair. However, the question also contains the additional information that the questioner swims five days a week and has black 1https://answers.yahoo.com/ Table 1: Example of question-summary pair Question Text: I’m a swimmer for my school swim team and I practice two hours a day, five days a week. I would like to dye my hair black (it is dark brown now) but I am wondering whether the chlorine will stripe it. Will it or will it not ? Summary: Will the chlorine stripe my hair ? hair. Although such information can sometimes be important for finding the exact answer, it is often peripheral when we want to grasp what is being asked. Summarizing a question, which can often be lengthy, helps respondents understand the ques- tion. The task of question summarization has not been studied yet and is worth being explored. In this work, we focus on a CQA site and examine the characteristics of the question summarization task with a CQA dataset as a case study. Specifically, we first examine CQA data consisting of pairs of a question text and its title, which we refer to as “question-title” pairs. We then propose a method for creating pairs of a question and its summary, which we refer to as “question-summary” pairs, out of the CQA data. We also propose methods for question summarization and describe an empirical evaluation we conducted. Approaches used in generic summarization tasks are often classified into two different …

Content truncated to 500 words