|
With the rapid development of internet, Q&A system has become a neccessary tool to get all kinds of knowledge and information. However, rapidly changing of information request Q&A system return accurate answers to users in time, which is important problem to be solved. Currently, Q&A systems already can find useful answers to a query by retrieving massive information resources, but some queries with time constraints still can not be answered well. If temporal factors can be considered into the system, including that analysis temporal information about the query and set temporal constraints for searching answers, then the accuracy of final answers would be increased in some extent.In order to improve the accuracy of answers, this paper classifies queries with time, and resolves the time-domain of query. Main tasks are as follows: (1)We take queries from online QA Community as corpus, and evaluate the consistence of labeling results with Fleisss’Kappa. (2)We do feature extraction with words, syntactic relation and semantic. (3)In the classification, AdaBoost Algorithm is used to balance the corpus, and combined with three machine-learning algorithms: Decision-tree, Na?veBayes and SVM to do classification test, the average value of F is 83%. (4)After classification, this paper proposes the method of resolving the time-domain of query, with which we can filter the search results, and the answering precision is increased by nearly 7 percentage points during processing of evaluating the system. Finally, with constraints of time-domain, the accuracy of results from baidu and soso are also increased.
|