一种融合多种语义特征的中文问题分类方法  

Study on Question Classification Approach Mixing Multiple Semantics Characteristics

作　　者：段利国[1] 陈俊杰[1] 牛彦清[1] 

机构地区：[1]太原理工大学计算机科学与技术学院,太原030024

出　　处：《太原理工大学学报》2011年第5期494-498,共5页Journal of Taiyuan University of Technology

基　　金：国家自然科学基金项目(60970059);山西省国际科技合作计划基金资助项目(2009081022)

摘　　要：针对中文问题分类方法中提取语义信息不准确和特征向量维数过高导致处理速度过慢的问题,提出了一种融合多种语义特征的问题分类方法。借助HowNet,兼顾问句的句法和语义信息,选取问题疑问词、核心词的主要义原、命名实体、名词单/复数等四种分类特征,并在义原的提取过程中加入词义消岐技术,对事实疑问句进行分类。在某高校信息检索研究室的中文问题集上进行实验,实验结果证明了该方法的有效性,大类准确率92.82%,小类准确率84.45%,取得了较好的效果。Amimed at the problems of inaccuracy and jogging speed resulted from too many characteristic vector dimensions in extracting semantic information during Chinese question classification, a kind of question classification method mixing multiple semantics characteristics was put forward. By taking account of sentence structure and semantics information and joining ambiguity eliminating technology during sememe extracting, the four classification characteristics including question interrogative, the main sememe of nuclear keywords, named entities and noun odd/plural forms were selected to classify the fact question sentence with the help of HowNet. The classification method was tested on Chinese question set of the Information Retrieval Lab in a university. Results show that the accuracy of broad heading and subclass achieved 92.82% and 84.45 %, respectively.

关 键 词：问题分类 疑问词 义原 命名实体 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]