Generating Synthetic Comparable Questions for News Articles

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 742–751, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Generating Synthetic Comparable Questions for News Articles Oleg Rokhlenko Yahoo! Research Haifa 31905, Israel olegro@yahoo-inc.com Idan Szpektor Yahoo! Research Haifa 31905, Israel idan@yahoo-inc.com Abstract We introduce the novel task of automati- cally generating questions that are relevant to a text but do not appear in it. One mo- tivating example of its application is for increasing user engagement around news articles by suggesting relevant compara- ble questions, such as “is Beyonce a bet- ter singer than Madonna?”, for the user to answer. We present the first algorithm for the task, which consists of: (a) of- fline construction of a comparable ques- tion template database; (b) ranking of rel- evant templates to a given article; and (c) instantiation of templates only with enti- ties in the article whose comparison un- der the template’s relation makes sense. We tested the suggestions generated by our algorithm via a Mechanical Turk ex- periment, which showed a significant im- provement over the strongest baseline of more than 45% in all metrics. 1 Introduction For companies whose revenues are mainly ad- based, e.g. Facebook, Google and Yahoo, increas- ing user engagement is an important goal, leading to more time spent on site and consequently to in- creased exposure to ads. Examples for typical en- gaging content include other articles for the user to read, updates from the user’s social neighborhood and votes or comments on videos, blogs etc. In this paper we propose a new way to increase user engagement around news articles, namely suggesting questions for the user to answer, which are related to the viewed article. Our motivation is that there are questions that are “irresistible” because they are fun, involve emotional reaction and expect simple answers. These are comparative questions, such as “is Beyonce a better singer than Madonna?”, “who is better looking, Brad Pitt or George Clooney?”, “who is faster: Superman or Flash?” and “which camera brand do you prefer: Canon or Nikon?” Furthermore, such questions are social in nature since users would be inter- ested in reading the opinions of other users, similar to viewing other comments (Schuth et al., 2007). Hence, a user that provided an answer may return to view other answers, further increasing her en- gagement with the site. One approach for generating comparable ques- tions would be to employ traditional question gen- eration, which syntactically transform assertions in a given text into questions (Mitkov et al., 2006; Heilman and Smith, 2010; Rus et al., 2010). Sadly, fun and engaging comparative questions are typically not found within the text of news articles. A different approach would be to find concrete relevant questions within external col- lections of manually generated comparable ques- tions. Such collections include Community-based Question Answering (CQA) sites such as Yahoo! Answers and Baidu Zhidao and sites that are spe- cialized in polls, such as Toluna. However, it is highly unlikely that …

Content truncated to 500 words