The Research On Question Understanding Technology Of Chinese Question Answering System

Question Answering (QA) system is an advanced form of the information retrieval system. It can provide an accurate and concise answer to a natural language query. The reason that QA technology thriving is the demand that people hope find the information quickly and accurately. QA system is paid more attention and it is a promising research direction in the field of Artificial Intelligence and Natural Language Processing.The Question Understanding (QU) technology of QA is studied in this paper. QU is the foundation of QA. Accurate results can be returned only by analyzing and understanding the questions properly. Question classification is the core of the QU. Based on deep study on the current QU technology, this paper makes several works as follows:1. In order to understand questions deeply and grasp the questions semantic, a semantic describing model named Event Frame Model is advanced. And we take HowNet event sememes and attribute sememes as the central part of the Event Frame.2. On basis of the idea of "question's type represents question intention", question intention based question classification is proposed in this paper. And in the Event Frame system, the formal question intention definition is shown as follow. Namely question intention is represent with the core of the event frame, as well as the answer type and the semantic role of the query-focus.3. The study of automatic recognition of the question intention. A two-level Maximum Entropy (ME) model is proposed to recognize the answer type, whose result is 2% higher than other approaches on the same training-testing corpus. A Hownet confidence evaluation based method to extract the central part of the Event Frame is advanced, which could extract the same central part in questions which have the same intention but in different expression forms. ME model is used for the query-focus Semantic Role Labeling (SRL). In SRL process, feature selection and optimization is studied deeply and modified Mutual Information model based feature selection method and a disassembled represent form for the Event Frame feature isproposed in this paper. Experiment result shows that the SRL precision is enhanced through these two methods.4. A QU sub-system which can be used for Pattern Based QA system is built. And the performance of the system is evaluated through the experiment.