基于HM-SVMs的问句语义分析模型

A QUESTION SEMANTIC ANALYSIS MODEL BASED ON HM-SVMs

由于传统的问句语义分析主要针对事实类的简单问句,而对于面向开放域的复杂问句缺少有效的语义分析方法。针对这种情况,提出一种新的问句语义分析模型。该模型将问句从文字空间映射到结构化的语义空间,实现问句的语义分析和表示。通过标注问句中的语义信息,模型实现问句分类、问句主题识别、限制信息识别三项分析工作。使用隐马尔科夫支持向量机(HM-SVMs)序列化标注工具实现了模型的自动标注,取得了86.7%的准确率。实验结果表明,HM-SVMs在标注准确率和效率上好于MEMM、CRF、M3N等模型,达到了预期效果。

Traditional question semantic analysis mainly focus on simple questions in regard to category of facts,but lacks effective semantic analysis method for open field-oriented complex questions. In view of this,we present a new question semantic analysis model. The model maps questions from text space onto a structured semantic space,and achieves semantic analysis and expression of questions. By annotating semantic information in questions the model implements three kinds of analysis works of questions classification,question topic identification and restrictive information identification. We employ hidden Markov support vector machines（ HM-SVMs）,a serialisation annotation tool,to realise the automatic annotation of the model,and reaches an accuracy of 86. 7%. Experimental results show that HMSVMs is better than MEMM,CRF,M3 N and other models in annotation accuracy and efficiency,and achieves the desired effect.