Research On Chinese Question Answering System Based On Semantic Role Labeling

Question answering is the next generation of search engine which is related to natural language processing, information retrieval and etc. The system allows the user to ask questions by using natural language, and it can also provide an accurate and concise answer. The reason that question answering technology thriving is the demand that people hope of finding the information quickly and accurately.Adding semantic information to question answering system is an inevitable trend for researching. However, the lower accuracy of Chinese semantic role labeling will make the question answering system's performance worse. That is why most systems only use words, part of speech and syntactic relations in question answering system. But, these features are very difficult to improve the accuracy of the question answering system now. For this reason, this paper presents a semantic relation validation method, and make it serve for the Chinese semantic role labeling. This method improves the accuracy rate of semantic role labeling. Then semantic roles of sentences are used in answer extraction of the question answering system, and it improves the system's accuracy.Question answering system is composed of three parts which is question analysis, information retrieval and answer extraction. This paper mainly focuses on question analysis and answer extraction.Question analysis includes question classification, keyword extraction and keyword expansion. This paper extract mainstay and question word offshoot of the question sentences as features. Then we employ Bayesian classifier to classify these questions.Answer extraction is the most crucial part for question answering system. The quality of answer extraction directly determines the performance of the question answering system. Its core issue is how to extract the answer from the candidate answer sentences accurately. This paper combines the question and answering sentences matching based on semantic roles, with the answer classification based on statistical method. The experiment results show that the method presented in this paper has a good effect.