Research On Chinese Question Generation And Query Reduction With QueryLog

With the explosive growth of Internet, How can we make Internet users get information easy become a key problem focused by academia and Search Engine companies. Users submit different kinds of queries on Search Engine when they want to know something. Analysizing queries submitted by Search users is very useful for us to understand what they want. It is important and promising for us to research on this techonology.Search users'querying and clicking behavior are stored in QueryLog. QueryLog is used on user behavior analysizing and improving search result. This paper focuses on the task of Auto Question Generation and Query Reduciton, all the corpus we use are all from Internet Search Engine QueryLog.Firstly, we propose a method automatically generate question from query for community-based question and answering (cQA). The use of cQA makes users get high quality information. Users can ask and answer questions, besides evaluate other users'answer. First of all, we get templates for question from QueryLog, and design a way to match query and templates. When user submits a new query, we find a best template for the query, and then insert the query into the template. The precision of 1-best generated question is 67%, outperforms cQA search engine Baidu Zhidao.Secondly, we propose a Chinese Query Reduction method. Queries are become longer, and the search result of long queries is not well. We transform Query Reduction into Classification problem with SVM and Sequence Labeling problem with CRFs to identify whether a word is key term in verbose queries. Experimental results show Sequence Labeling method preforms better than Classification method. We also add effective features by analysizing the effect of different features.