解答小学数学应用题是培养小学生问题解决能力的重要方式，智能辅导系统高性能地交互学习环境可以有效的辅导小学生解答数学应用题。许多研究者开发出了相关系统用以促进小学生应用题解题能力的提高。但受现有自然语言处理等技术的局限，目前已有系统的自动化程度不高。提高理解和表征应用题的效果对于提高智能辅导系统自动化程度的具有重要的作用。因此，小学数学应用题语句语义理解的研究，对于提高系统的自动化程度具有重要作用。应用题语句可以分为情境句和非情境句。由于非情境句带有数据信息，因此非情境句在整个应用题语句中占有重要地位。非情境句又可以分为赋值语句和关系语句。“有”字句是一种基本赋值语句，比较句是一种重要的关系语句，且“有”和“比”是应用题中出现频率最高两个词。因此实现“有”字句和比较句的语义理解，对于实现其它非情境句的理解，以及所有应用题语句的理解都具有重要的作用。本研究总结了“有”字句和比较句的句式，提出了识别两种语句的算法，同时总结了存储两种语句重要信息的语义框架。对于“有”字句，本研究将其分为简单“有”字句和复杂“有”字句，并采用基于规则的方法进行研究。对于简单“有”字句，提出了5个算法来填充语义框架；对于复杂“有”字句，结合句中的关键字以及简单“有”字句的处理算法进行处理。最后利用实验验证算法的有效性，并采用三个指标进行评价，实验结果理想。对于比较句，本研究采用条件随机域模型的一个工具包CRF++进行处理。首先选择标记集、确定特征和特征模板；然后人工标注训练语料和测试语料；使用训练语料训练模型，并利用训练出的模型对测试语料进行标注，从而识别出测试语料中比较句的各种成分，填充语义框架；最后将测试语料的人工标注结果与模型标注结果进行对比，采用四个指标进行评价，标注结果较为理想。

Solving word problems is an important method to train pupils the ability of problem solving. The interactive learning environment of Intelligent Tutoring Systems can tutor pupils to solve word problems effectively. Many researchers have developed related systems to help students. Due to the limitations of existing Natural Language Processing technology, the automation degree of existing systems is not high. The first step of increasing the degree of automation is to understand and represent word problems sentences better. In conclusion, it is very important to achieve semantic understanding of word problems sentences.Sentences of word problems can be divided into situational sentences and non-situational sentences. Non-situational sentences carry data information, so the understanding of them is very important for all the sentences. Besides, non-situational sentences can be divided into assignment statements and relational statements. "You"-sentence is a kind of basic assignment statements, comparative sentence is an important kind of relational statements, and "You" and "Bi" are two words with highest frequency of occurrence in word problems. In conclusion, the semantic understanding of "You"-sentence and comparative sentence can promote the understanding of other non-situational statements and even all word problems statements.This study summarizes the sentence patterns of "You"-sentence and comparative sentence, proposes three algorithms to identify these two statements, and extracts semantic frameworks of them. For "You"-sentence, this study uses the rule-based approach, and proposes five algorithms to extract the important information of the simple "You"-sentence and fill out the related framework. And then combined with the keywords of complex "You"-sentence, it achieves the semantic understanding of complex "You"-sentence. Finally, it uses experiments to verify the efficiency of the above algorithms, the results are ideal. For comparative sentence, this study uses conditional random fields model to process it. First this study selects the tag set, extracts features and determines the feature templates. And then it marks training corpus and test corpus manually, trains the model, and uses the trained model to label test corpus. Finally, it uses the overall accuracy, precision, recall and F value to evaluate the results of labeling.