摘要
该文提出一种支持向量机(support Vector Machines,SVM)和粗糙集(Rough Set,RS)相结合的巾文机构名称短语识别方法。该方法借助词的基术语义搭配关系表示短语的构成规则,并通过粗糙集属性约简的方法自动学>J 到机构名称构成规则的无冗余集。识别时,首先寻找到与这些规则匹配的词串作为候选机构名,然后结合候选机构名以及其上下文词的语义特征,利用SVM分类器判断该候选是否是真正的机构名称。这种方法对1617万字人尾日榴语赳开卉々jIj=『Il请的F信钋剐诀到R,f16%.
A method to identify Chinese organization names by utilizing SVM (Support Vector Machines) and RS (Rough Set) is provided. Forming rule of organization name is defined based on semanteme collocation relation, and then the un-redundancy set of rough forming rules can be learned by employing attribute reduction in RS automatically. A chain of words matching forming rule is selected first as candidate, then a SVM classifier discern whether a candidate is real organization name according to candidate semanteme and its contextual semanteme while recognizing. Results of open testing achieve F-measure 82.06% in 16.17 million words news based on this project separately.
出处
《电子与信息学报》
EI
CSCD
北大核心
2006年第5期895-900,共6页
Journal of Electronics & Information Technology
基金
国家自然科学基金(60175020)
国家863计划(2002AA117010-09)资助课题
关键词
模式识别
SVM
特征选择
语义
粗糙集
语义搭配
Pattern recognition, SVM, Feature selection, Semanteme, Rough Set(RS), Semanteme collocation