Research And Implementation Of Answer Extraction For Open Domain Question Answering System

Search engine requires professional keywords combination and returns list of relevant documents.On the contrary,automatic question answering system allows users to ask questions in natural language,and returns accurate and specific answers,which not only helps users express their search intention more clearly and easily,but also may get a higher efficiency in seeking for information they need.Therefore,automatic question answering system is supposed to meet the users' needs of information retrieval better.Consequently,question answering system has become another hot spot within information retrieval domain,especially open domain question answering system based on the internet,which is of significant research value and broad application prospect.This thesis focuses on the critical technologies in the process of open domain question answering system,such as the question keywords extension,information extraction and answer quality assessment.The main research achievements are as follows:First,a keyword extension model is designed.Based on the characteristics of the search engine and the demands of question answering system,and with the help of the thesaurus of WordNet,a keyword extension model is designed to expend the query reasonably.Second,an information extraction model based on text similarity is proposed,in which the text similarity is measured by combination of multi features,and the machine learning algorithm is utilized to extract the candidate answers.Third,an answer quality evaluation method based on similar support set is proposed,which takes the high quality answers of history similar questions as the support set and calculates the relevance between question and answer indirectly.Reducing the impact of the large direct semantic distance between question and answer,this model improves the efficiency of answer quality evaluation.Fourth,an open domain question answering system for internet users is designed and implemented,applying the models and technologies presented above.Experiments are also designed and taken to demonstrate the efficiency of each module and the availability of the system.