Question Identification In Chinese Micro-texts

Nowadays, social networks, such as Twitter, Facebook and Sina Weibo, have become one part of modern life. Micro-blogs are no longer social medias which let user publish "What is happening"s but also become online questioning platforms where users seek information. More and more people are asking questions by social networks, and they ask questions for two reasons. First, there are some questions that they cannot find the answers by traditional search engines. Second, they prefer to trust the answers from their friends or experts.In this paper, we focus on automatically identifying questions (the micro-texts which are asked to seek information or help) from Chinese micro-texts. On one hand, word segmentation is usually the first step of Chinese text processing, and it could cause error propagation. Since there are many out-of-vocabulary (OOV) words in Chinese micro-texts, we propose a Chinese word segmentation method with abstraction on character levels which could improve the performance of OOV words. On the other hand, different from English, there are more Chinese question words and Chinese sentence is more complex. Due to the diversity of Chinese questions, we propose a learning-based method to recognize question with dependency tree pattern, which uses both lexical and syntactic information. In addition to identify the questions, we also distinguish whether the questions can be answered immediately with sufficient conditions.The experimental results on sampled two-day Chinese micro-texts show that our method is very effective for question identification. Compared with other method, our method achieves better performance. Besides, the performance of our method on sufficient question identification is much better, which shows our method is more effective on more complex tasks.