【作者】 李超；

【导师】 杨进才；

【作者基本信息】 华中师范大学 ， 计算机技术（专业学位）， 2017， 硕士

【摘要】 目前,在中文信息处理领域中,对字和词的研究已经趋于成熟,相应的研究成果得到了广泛地应用。复句在汉语句法中具有相当重要的地位,而复句关系词又是小句连接的重要手段,因此,对复句进行深入研究始终离不开复句关系词。复句关系词的研究是以关系词的自动识别为基础的,目前主流的方法是基于规则和统计模型相结合的方法。虽然现在对复句关系词的自动识别已经取得了较好的效果,但准确率还有待提高。原因之一是关系词识别是基于分词工具进行的,词语切分和词性标注的错误,往往会给识别结果造成一定的误差。针对这一问题,本文提出了一种基于规则和VSM相结合的复句关系词切分歧义的消解算法。本文首先利用中科院的NLPIR汉语分词系统对汉语复句标注语料库进行词语切分和词性标注的预处理,对预处理后的复句语料进行统计和分析,总结出复句关系词切分歧义字段的范围、性质和分布情况,根据切分歧义字段的表现形式对其进行归纳分类;其次,对切分歧义字段进行定量分析,抽象出特征规律,并将这些特征规律形式化为规则,利用总结的规则识别复句关系词的歧义字段;然后,根据VSM的特点,分别构造训练集和测试集,通过训练集利用CHI方法确定词性矩阵的大小,得到分类效果最优的上下文范围;最后,选择词性作为特征项,构造出词性矩阵和待消歧义字段的上下文向量并分别映射到n维空间中,通过计算上下文向量到每个词性向量的距离,以距离最近的词性向量所代表的词性类型来标注该歧义字段,从而达到整个消歧的目的。实验表明,基于规则和VSM的方法对复句关系词切分歧义的消歧取得了较高的正确率,消歧的正确率达到了 95.94%,表明本文提出的方法是可行的、有效的。更多还原

【Abstract】 In the field of Chinese information processing,word and expressions research has become mature these day.And the corresponding research results have been widely applied.The compound sentence in the Chinese syntax has a very important position.The relation word of compound sentence is an important means of connecting clause.Therefore,the study of compound sentence is still inseparable from the relation words.The study of compound sentence is based on the automatic recognition of relation words.The current mainstream is based on the combination of rules and statistical models.Although the automatic recognition of the compound sentence has been achieved some success,but the accuracy rate has yet to be improved.One of the reasons is the relation words identification is based on the word segmentation system.The error of automatic word segmentation and part-of-speech tagging will cause deviation for recognition results.To solve this problem,this paper proposes a method to eliminate the ambiguity of compound sentence words based on rules and vector space model(VSM).In this paper,we first used the NLPIR Chinese word segmentation system of the Chinese Academy of Science to pre-process the automatic word segmentation and part-of-speech tagging for Chinese compound sentence corpus,then counted and analyzed the pre-processing compound sentence corpus.And we summarized the scope,character and distribution of the compound sentence relational word.According to the manifestations of segmentation of ambiguous fields,we induced and classified it.Secondly,we quantitatively analyzed the segmentation ambiguous fields and extracted the characteristic patterns,formalized the characteristic patterns into rules,and identified the ambiguous fields of the compound sentence.Then,according to the characteristics of VSM,we constructed training set and test set.Through the training set,we determined the size of the part of speech matrix and got the optimal context of the classification effect by the CHI method.Finally,we chose part-of-speech as a quantitative weight strategy,constructed the part-of-speech matrix and the context vector of ambiguous field which is to be eliminated.And we mapped it to mufti-dimensional space,calculated the distance from the context vector to each part-of-speech vector.We use the nearest part-of-speech vector of the part-of-speech type mark the ambiguous field,so as to achieve the purpose of the whole disambiguation.Experiments show that the method based on rules and VSM has a high correct rate for disambiguation of relation words.Disambiguation of the correct rate reached 95.94%.It shows that the method proposed in this paper is feasible and effective.更多还原