句法分析是自然语言处理领域中应用前景非常广阔的一个研究方向.针对目前句法分析多数是从字、词的角度出发且存在诸多不足,提出了二、三元词模型相结合的句法规则层次化分析算法,并结合分词、词性标注以及句子组织信息之间的结合度来解决词元间优先合成的问题,同时利用句子成分之间的语法结构关系对词性、词序的影响,实现句法规则的层次化分析实验.实验结果表明,二元与三元词模型相结合的句法规则层次化分析算法相比于独立二、三元词模型,准确率和召回率分别提高了82.04％ 和80.83％,与现有基于二分结构句法分析的RN N-IN T算法和词汇化模型算法相比,准确率和召回率均有明显提升.

This thesis discusses the decomposition of natural language sentences based on the rules of order-modified syntax. By making a comprehensive use of the combination of part-of-speech tagging, syntactic structure and sentence organization information, we propose to apply the custom rules to achieve the decomposition of sentences. The visualization of the decomposition results under a certain condition of part-of-speech tagging and word order is realized, which provides a new idea for the development of syntactic analysis structure.

Text feature is an important category attribute of text. Feature extraction directly affects the accuracy of text classification. An improved feature weighting algorithm is proposed in this paper. The chi-square statistical method is applied to calculate feature weight, which improves the accuracy of extracting feature words of categories. The IDF calculation method is improved from the category concentration of keywords. This paper uses the TF-IDF before and after the improvement to extract feature and classify the same text data individually. The results show that the classification effect of the improved method is better than the traditional method.

首先,在句子组织信息之间的结合度及基于规则、词性和词序对句法分析系统影响的基础上,提出一种基于规则的语句分析识别算法,能在大量文本中快速识别出正确句式;其次,在基于语句分析识别算法的基础上,提出一种基于规则与句法合成的层次化语句分析识别算法,以提高层次化句式识别检错的精度.实验结果表明,该算法平均精确率和平均召回率分别为84.65%和77.15%,相比于只基于规则的语句识别算法分别提高了11.79%和14.48%,证明了规则与句法合成的层次化语句分析识别的可行性.