多领域自然语言问句理解研究

问句理解是问答系统的主要任务之一。现有的问句理解方法大多是针对简单句的,且侧重于某种句式结构的理解。提出一种多领域问句理解研究方法,其涉及领域包括人物类、电影类、音乐类、图书类、游戏类、应用类。首先基于CRF算法对问句进行分类和主体识别,然后使用谓词词典和句法分析识别出问句的谓词,最后提出一种谓词消歧方法来解决相同问句具有不同表达方式的问题。实验结果表明,在封闭测试中,所提方法的问句分类和主体识别的平均F-measure值分别为93.88%和92.44%,谓词识别和问句理解的平均准确率分别为91.03%和81.78%。因此,所做的工作基本能满足问句理解的需求。

Research on Multi-domain Natural Language Question Understanding

Question understanding is one of the main tasks of question answering system.Current question understan-ding methods aim to solve semantic understanding of simple sentences or specific structure sentences.The method proposed in this paper addresses multi-domain question understanding which includes people,movie,music,book,game,and application domains.Firstly,the question classification based on CRF algorithm and the subject recognition based on CRF algorithm approach are presented.And then the prediction dictionary and semantic analysis are applied to recognize prediction.Finally,the prediction disambiguation method is proposed to deal with the problem that prediction in question has different ways of expression.Experimental results show that the average F-measure value is 93.88% and 92.44% in question classification and semantic analysis experiments.The average accuracy is 91.03% and 81.78% in the prediction recognition and question understanding.Thus,the works in this paper can meet the needs of question understanding.