基于N元模型的维吾尔文文本分类技术研究  

Research N-gram based Uyghur text classification technique

作　　者：买买提依明.哈斯木 吾守尔.斯拉木 维尼拉.木沙江 努尔麦麦提.尤鲁瓦斯 

机构地区：[1]新疆大学信息科学与工程学院新疆多语种重点实验室,乌鲁木齐830046 [2]和田师范专科学校计算机科学系,新疆和田848000

出　　处：《计算机应用研究》2015年第7期1986-1988,2004,共4页Application Research of Computers

基　　金：国家"973"重点基础研究计划基金资助项目(2014CB340506);国家自然科学基金资助项目(61262063;61363063)

摘　　要：考虑到维吾尔文词干提取、词性标注等工具不够成熟和相关的开源资源很少的实际情况,提出了基于N元模型的维吾尔文文本分类技术。其特点是不需要任何自然语言处理工具,拼写错误率对分类结果的影响很低。在训练阶段分别提取字符级别的三元和四元模型构造不同规模的N元词典,在分类测试阶段分别用曼哈顿距离计算和骰子测量对文本进行分类。实验结果表明,当四元模型词典的规模为500时,使用骰子测量分类时性能最佳,平准准确率达到86.56%。Considering Uyghur stemming, POS tagging and other tools are not mature enough and there are a few open resources, this paper proposed N-gram based Uyghur text classification technique. The advantages were don＇ t needs any natural language processing tools and misspelling had low impact on text classification. In the process of learning phase extracted respectively character level tri-grams and quad-grams and constructed different scale N-gram profile, in the classification process respectively used Manhattan distance and Dice measure to classified text. The experimental results show that when quad gram profile size 500 and use Dice measure has best classification performance. The average accuracy rate reaches 86. 56%.

关 键 词：N元 文本分类 维吾尔文 N元词典 相似度 曼哈顿距离 骰子测量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]