周洲 侯开虎 姚洪发

摘  要： 随着计算机辅助教学，多媒体处理以及计算机网络技术的发展与成熟，目前已经有许多考试都采用无纸化考试，即机考的形式进行。采取电子化考试的优点在于考试可监控性强，考试效率高，考试标准化和程序化。并且对于选择题判断题这样的客观题自动化批改技术已经十分成熟，极大的缩减了改卷时间，提高了改卷效率。但是，由于受到自然语言理解的限制，至今没有比较完善的主观题自动批改系统对主观题进行批改和评分。本文采用TF-IDF及LSI两种模型作为文本分析模型，使用jieba中文分词工具进行文本预处理，使用Python语言实现该系统。通过考生答案与标准答案的语义相似度分析，对考生作答的主观题进行批改和评分。之后随机抽取5份大学考试中的考生试卷，使用该主观题自动评分系统进行测试，与改卷老师所给出的评分进行对比分析和说明。实验结果表明，本文所提出的主观题自动评分系统在一般情况下可以满足主观题自动评分的功能，是一种值得继续深入研究的可行方法。

关键词： 无纸化考试;主观题;自动评分;Python;TF-IDF;LSI

【Abstract】： With the development and maturity of computer aided instruction， multimedia processing and computer network technology， many examinations have been conducted in the form of paperless tests， that is， computer tests. The advantage of electronic examination is that it can be monitored well， the efficiency of examination is high， the examination is standardized and programmed. And for the multiple choice judgment questions such as automatic marking technology has been very mature， greatly reduced the time to correct the paper， improve the efficiency of paper correction. However， due to the limitation of natural language understanding， there is no perfect automatic subjective question marking system. In this paper， TF-IDF and LSI models are used as text analysis models， jieba Chinese word segmentation tools are used for text preprocessing， and Python language is used to implement the system. By analyzing the semantic similarity between the test answers and the standard answers， the subjective questions are corrected and graded. Then 5 college examination papers were randomly selected and compared with the real scores of teachers. The experimental results show that the automatic scoring system of subjective questions proposed in this paper can meet the function of automatic scoring of subjective questions under general circumstances and is a feasible method worthy of further study.

【Key words】： Paperless test; Subjective questions; Automatic scoring; Python; TF-IDF; LSI

0  引言

主观题自动评分系统，是指使计算机能够“理解”人类自然语言，在此基础上，对自然语言构成的句子，段落进行评价，评分的系统。虽然近几年人工智能、大数据等技术日益成熟，但是国内应用计算机技术进行的考试还停留在只有选择题可以进行自动评分的程度，对于填空、简答、论述等题目尚未出现比较实用可靠的自动评分技术，而在一般情况下，批改试卷的老师往往需要在极短的时间下对一份试卷中的一个或几个题目进行打分。……