摘要
针对英文现在分词词性标注这一特定问题存在的难点分析了隐马尔可夫模型(HMM)的不足,提出了贝叶斯决策树模型。对一个已经标注好的语料库进行统计,运用决策树C4.5算法从单边条件和双边条件两个方面对英语现在分词的三种词性进行合理的分类消歧。对于双边条件下仍然存在歧义的情况,用贝叶斯最小风险对决策树改进,用标注好的语料库对模型进行训练。最后,采用一个未经过标注的语料库进行测试,取得了非常好的效果,证明了模型的优越性。
Concerning the difficulties in part-of-speech tagging in English present participle, the authors analyzed the drawbacks of Hidden Markov Models (HMM) and proposed Bayesian decision tree model. Firstly, the tagged corpus was calculated and CA. 5 in decision tree was used for proper classification and disambiguation of the three classes of present participle. Then, the decision tree was improved by Bayesian least risk. At last, an untagged corpus was used to test the model and the result is very good, which proves the superiority of the model.
出处
《计算机应用》
CSCD
北大核心
2009年第9期2571-2574,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(60773169)
关键词
分类
消歧
贝叶斯决策树
隐马尔可夫模型
classification
disambiguation
Bayesian decision tree
Hidden Markov Model (HMM)