BJUT at TREC 2016: LiveQA Track

The paper presents the BJUT’s liveQA system participating the TREC 2016. The Trec LiveQA track continues to use the last year’s instruction, requiring that the system is able to answer the questions which had not been solved in one minutes based on Yahoo! Answers. Our work: (1) The key words are abstracted from the questions, so that more relevant questions will be returned. (2) The system searches in a larger scope on Yahoo! Answers to find the most accurate answers. (3) The answers should be detect whether they are more suitable for answering the given questions. The experiment results are presented at the end of the paper. Introduction The automated question answering (QA) track, which has been one of the most popular tracks in TREC for recent years, has focused on the task of automatically answering questions posed by humans in a natural language. The track primarily dealt with factual questions, and the answers provided by participants were extracted from a collection of news articles. While the task evolved to model increasingly realistic information needs, addressing question series, list questions, and even interactive feedback, a major limitation remained: the questions did neither come from real users, nor in real time(Robertson and Walker 1997; Mikolov et al. 2013). The Trec LiveQA track mainly aims at providing the automatic answers for questions posed by humans in a natural language. There is also an additional demand that extracts the keywords from the question. This track revives and expands the QA track, focusing on live question answering for real-user. Real user questions, extracted from the stream of most recent questions submitted on the Yahoo Answers (YA) site that have not yet been answered by humans, will be sent to the participant systems. The systems will provide an answer in real time. The list of YA categories is limited to a certain range, which includes Arts & Humanities, Beauty & Style, Health, Home & Garden, Pets, Sports and Travel. The question will be provided every minute for a whole day. The returned answers is restricted to 1000 characters and will later be judged by TREC editors on a 5-level Likert scale. This paper introduces our liveQA system which we use to accomplish the Trec LiveQA track task answering the questions in real time. Since the questions are all from Yahoo Answer, we assume that the questions input into the system have been asked by other people previously, and these similar questions have already had best answers. So we transfer the task from answering the questions to choosing the best answers by similar questions. We don’t use any search engine, because we think the answer in Yahoo! Answers is more general.