Extractive Headline Generation Based on Learning to Rank for Community Question Answering

Proceedings of the 27th International Conference on Computational Linguistics, pages 1742–1753 Santa Fe, New Mexico, USA, August 20-26, 2018. 1742 Extractive Headline Generation Based on Learning to Rank for Community Question Answering Tatsuru Higurashi† Hayato Kobayashi†‡ Takeshi Masuyama† Kazuma Murao† †Yahoo Japan Corporation ‡RIKEN AIP {thiguras,hakobaya,tamasuya,kmurao}@yahoo-corp.jp Abstract User-generated content such as the questions on community question answering (CQA) forums does not always come with appropriate headlines, in contrast to the news articles used in various headline generation tasks. In such cases, we cannot use paired supervised data, e.g., pairs of articles and headlines, to learn a headline generation model. To overcome this problem, we propose an extractive headline generation method based on learning to rank for CQA that extracts the most informative substring from each question as its headline. Experimental results show that our method outperforms several baselines, including a prefix-based method, which is widely used in real services. 1 Introduction Community question answering (CQA) is a service where users can post their questions and answer the questions from other users. The quality of a CQA service is inevitably linked to how many questions are answered in a short amount of time. To this end, the headline of a posted question plays a key role, as it is the first thing users see in a question list or push notification on a smartphone. However, questions in a CQA service do not always have appropriate headlines because the questions are written by various users who basically do not have any specialized knowledge in terms of writing such content, in contrast to news articles written by professional editors. In fact, the biggest CQA service in Japan, Yahoo! Chiebukuro1, does not even provide an input field for headlines in the submission form of questions, as general users do not have enough patience and tend not to post questions if even just one required field is added to the form. This service alternatively uses the prefix of a question as its headline as in Figure 1(a), where the headline “はじめまして。よろしくお願いします。仕事… (Nice to meet you. Thank you in advance. At work ...)” is created from the prefix of the content. Obviously, this headline is uninformative because (a) Example of posted question. (b) Example of push notification. Figure 1: Examples of (a) posted question and (b) push notification. This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http: //creativecommons.org/licenses/by/4.0/ 1https://chiebukuro.yahoo.co.jp/ 1743 of the lack of actual content, which is related to an initial unrequited love of a woman in the workplace. Figure 1(b) shows how ineffective such an uninformative headline is for a smartphone push notification, where users would have practically zero motivation to click on it and answer since they cannot imagine what kind of question is being asked. This negative effect has been confirmed on a commercial CQA service (Section 2). In this work, we take an extractive approach for improving uninformative headlines. Although there have recently been many studies on abstractive headline generation (as described in Section 6), …

Content truncated to 500 words