基于SVM的初等数学问题自动分类的研究与应用

发布时间：2018-10-29 21:03

【摘要】：众所周知,随着计算机信息技术的迅速发展,信息技术已经在我们生活的方方面面有所应用。在教育领域,人们的目光也逐渐从线下老师辅导以及手工批阅试卷,解题转向了基于人工智能的互联网智能教育。这种新型的数学教育理念实现的一个重要前提就是对文本自然语言进行转换,通俗的讲就是把人类理解的数学语句转化为事先定义好的计算机存储知识,来供计算机进行下一步的处理。这些处理主要有解题,以及全流程判卷等等。这一前提也可以称为自然语言处理过程。而分类问题又是自然语言处理的过程中的主要问题。本文主要分为两个部分,第一部分是对初等数学问题文本进行分词,以及词性标注和命名实体识别。第二部分是基于SVM对初等数学问题文本进行题型分类,进而根据不同的类别转换成计算机推理所需的表现形式。在英文表达中,每个单词之间都是有空格的,但是中文则不同,所有的字符都是连接在一起的,所以需要对中文的文本进行分词。但是数学表述中包含了较多的有特定含义的符号,所以通用的分词方法行不通。因此需要针对数学表述构造专门的分词器。同样,数学语言表达中的实体和普通语言表达的实体不同,普通语言的实体更多是时间,地点,姓名等。而在数学表达中,包含重要信息的实体往往是数学名词,比如三角形,以及各方程等等。因此需要针对初等数学方向定义专门的命名实体,然后进行提取。本论文采用条件随机场来进行命名实体标注。初等数学问题中涉及的类型有很多,想要对初等数学问题进行自动求解,首先要做的是把问题进行分类,然后根据不同的类别调用相应的求解方法。对经过命名实体模型标注的初等数学问题文本进行文本预处理,包括去停用词,建立词袋模型。在本论文中,通过卡方统计量来实现文本特征向量的选择。这样使用特征向量通过选择降维技术能很好的减少计算量,还能维持分类的精度。最后,根据本文提出的方法,使用支持向量机实现了一套对初等数学问题进行命名实体提取并且对题目进行分类的系统。该系统可以准确标注命名实体,为后面解题等处理提供知识表示,同时有效的题目分类可以为后面解题或者判卷做推理剪枝。
[Abstract]:As we all know, with the rapid development of computer information technology, information technology has been applied in every aspect of our life. In the field of education, people's eyes have gradually shifted from offline tutoring and manual marking of examination papers to intelligent Internet education based on artificial intelligence. One of the important prerequisites for the realization of this new concept of mathematical education is to transform the text into natural language. In popular terms, it is to convert mathematical statements understood by human beings into pre-defined computer storage knowledge. To allow the computer to handle the next step. These processing mainly have the solution, as well as the whole flow judgment paper and so on. This premise can also be called natural language processing process. Classification is the main problem in the process of natural language processing. This paper is mainly divided into two parts. The first part is the participle of elementary mathematical problem text, as well as part of speech tagging and named entity recognition. In the second part, the paper classifies the text of elementary mathematics problem based on SVM, and then transforms it into the representation of computer reasoning according to different categories. In English, there is a space between each word, but Chinese is different, all characters are connected together, so the Chinese text should be partitioned. However, mathematical expressions contain more symbols with specific meanings, so the general participle method is not feasible. Therefore, it is necessary to construct a special participle for mathematical expression. Similarly, the entities expressed in mathematical language are different from those expressed in common language. The entities of common language are more time, place, name and so on. In mathematical expressions, the entities that contain important information are often mathematical nouns, such as triangles, equations and so on. Therefore, it is necessary to define a specific named entity for the primary mathematical direction and then extract it. In this paper, conditional random fields are used to label named entities. There are many types involved in elementary mathematics problems. In order to solve elementary mathematical problems automatically, the first thing to do is to classify the problems and then call the corresponding solving methods according to different categories. The text preprocessing of primary mathematical problem text tagged by named entity model includes deactivating words and establishing word bag model. In this paper, chi-square statistics are used to select text feature vectors. In this way, the feature vector can reduce the computational cost and maintain the classification accuracy by selecting dimensionality reduction. Finally, according to the method proposed in this paper, the support vector machine (SVM) is used to implement a system for extracting named entities from elementary mathematical problems and classifying them. The system can accurately label named entities and provide knowledge representation for later problem solving and so on. At the same time, effective topic classification can be used as inference pruning for later problem solving or marking.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1;O12

【参考文献】