摘要: 针对问答题类文字描述性主观题机器阅卷的复杂性和困难性,提出一种用于机器阅卷的两级相似度计算算法。综合考虑答案的关键词、句子语法和语义信息,并结合分数微调规则设计算法。实验结果表明,该算法在词语级系数α取值约0.7时,阅卷系统具有最低的无效阅卷比例和较快的速度,符合人工阅卷的要求。
关键词:
相似度计算,
向量空间模型,
机器阅卷,
自然语言处理,
依存文法,
规则
Abstract: According to answers’ keywords, sentence grammar semantic informations and adjusting rules, this paper proposes a two level similarity computation algorithm which integrates the advantage of word level and sentence level similarity computation. It focuses on questions of text descriptive subjective examinations complexities and machine marking difficulties. Experimental results show that the system can obtain the minimum invalid marking ratio and fast speed, while word level coefficient α value is about 0.7. This algorithm complies with the effect of manual marking better.
Key words:
similarity computation,
vector space model,
machine marking,
natural language processing,
dependency grammar,
rule
中图分类号:
秦学勇, 张润梅. 两级相似度计算在主观题机器阅卷中的应用[J]. 计算机工程, 2012, 38(11): 274-276,280.
QIN Hua-Yong, ZHANG Run-Mei. Application of Two Level Similarity Computation in Subjective Machine Marking[J]. Computer Engineering, 2012, 38(11): 274-276,280.