摘要
为了解决自然语言处理领域中的一词多义问题,本文提出了一种利用多种语言学知识和词义消歧模型的半监督消歧方法.首先,以歧义词汇左、右邻接词单元的词形、词性和译文作为消歧特征,来构建贝叶斯(Bayes)词义分类器,并以歧义词汇左、右邻接词单元的词形和词性作为消歧特征,来构建最大熵(maximum entropy,ME)词义分类器;其次,采用Co-Training算法并结合大量无标注语料来优化词义消歧模型;再次,进行了优化实验,在实验中,使用SemEval-2007:Task#5的训练语料和哈尔滨工业大学的无标注语料来优化贝叶斯分类器和最大熵分类器;最后,对优化后的词义消歧模型进行测试.测试结果表明:与基于支持向量机(support vector machine,SVM)的词义消歧方法相比,本文所提出方法的消歧准确率提高了0.9%.词义消歧的性能有所提高.
To solve the problem of a word having multiple meanings in the natural language processing(NLP)field,a semi-supervised disambiguation method,that uses a range of word sense disambiguation(WSD)models and linguistic knowledge has been proposed in this paper.First,words,parts of speech and translations were used as discriminative features,which were extracted from word units adjacent to the left and right of an ambiguous word.A word sense classifier was constructed using a Bayes model,following which a word sense classifier based on a maximum entropy(ME)model was constructed.Second,a Co-Training algorithm,based on a multitude of unannotated corpora,was adopted to optimize the WSD model.Third,optimization experiments were conducted in which training corpus in SemEval-2007:Task#5 and a large number of unannotated corpora from Harbin Institute of Technology were applied to optimize the Bayesian classifier and the maximum entropy classifier.Finally,the optimized WSD model was tested.Test results demonstrate an increase in the disambiguation accuracy of the proposed method by 0.9%compared to WSD models based on support vector machines,thereby exhibiting an improvement in WSD performance.
作者
张春祥
徐志峰
高雪瑶
ZHANG Chunxiang;XU Zhifeng;GAO Xueyao(School of Software and Microelectronics,Harbin University of Science and Technology,Harbin 150080,China;School of Computer Science and Technology,Harbin University of Science and Technology,Harbin 150080,China)
出处
《西南交通大学学报》
EI
CSCD
北大核心
2019年第2期408-414,共7页
Journal of Southwest Jiaotong University
基金
国家自然科学基金资助项目(61502124
60903082)
中国博士后科学基金资助项目(2014M560249)
黑龙江省自然科学基金资助项目(F201420
F2015041)
关键词
自然语言处理
词义消歧
最大熵
贝叶斯分类器
natural language processing
word sense disambiguation
maximum entropy
Bayesian classifier