论文部分内容阅读
答案抽取是问答系统中的核心内容,问题及答案句的句法和语义充分理解是找出答案的关键.由于中文自然语言句法语义分析复杂,人工提取特征难度较大、主观性较强,使之成为中文问答系统的研究重点和难点.为此本文提出利用深度学习的思想主动学习候选答案深度特征,将答案抽取问题转化为特征学习与分类问题.即用词向量表征问题句和答案句,通过长短时记忆神经网络主动学习其深层语义相关,借助依存句法树分析句法结构特征,构造深度神经网络学习问题句、答案句和候选答案之间的内在关联信息.实验表明,该方法在不需要制定繁琐句法语义特征的条件下,仍具有较好的答案抽取性能,MRR值达到0.71.
The extraction of answers is the core of the question answering system, and the full understanding of the grammar and semantics of the questions and answers is the key to find out the answers.Because of the complex syntactic and semantic analysis of the Chinese natural language, it is more difficult and more subjective to extract the features manually Become the focus and difficulty of the Chinese question answering system.Therefore, this paper proposes the use of deep learning initiative to learn the depth of the candidate features, the answer extraction problem into feature learning and classification problems that use word vectors to characterize the question and answer sentences, through The short and long-time memory neural network takes the initiative to learn the deep semantic correlation, constructs the deepness neural network to learn the internal correlation information between the question sentence, the answer sentence and the candidate answer by analyzing the syntactic structure features with the aid of the syntactic tree.Experiments show that this method does not need to be developed Under the conditions of cumbersome syntactic and semantic features, it still has better performance of extracting the answers, and the MRR reaches 0.71.