基于相似度组合的主观题评分方法研究  

Research on Subjective Question Scoring Method Based on Similarity Combination

作　　者：肖灵云 刘军库 XIAO Lingyun;LIU Junku(School of Intelligent Manufacturing,Cunjin College of Guangdong Ocean University,Zhanjiang 524000,China)

机构地区：[1]广东海洋大学寸金学院智能制造学院,广东湛江524000

出　　处：《贵州大学学报（自然科学版）》2021年第5期64-68,共5页Journal of Guizhou University:Natural Sciences

基　　金：广东省自然科学基金资助项目(2018A0303130076)。

摘　　要：针对文本、语义、关键词等方法计算主观题相似度的不足,提出一种基于相似度组合的方式构建主观题评分模型。该模型可根据评分逻辑,灵活地调整阀值,综合Word2vec算法、Doc2vec算法和TF_IDF算法的优势,提高主观题的相似度。实验表明,与人工评分对比,提出的评分模型得到的评分方差及偏差率都较小,稳定性较好,更贴近人工评分结果。Aiming at the shortcomings of calculating the similarity of subjective questions by text,semantic and keyword methods,a subjective question scoring model based on similarity combination was proposed.The model can flexibly adjust the threshold according to the scoring logic,and integrate the advantage of Word2vec algorithm,Doc2vec algorithm and TF_IDF algorithm,it can improve the similarity of subjective questions.Compared with manual scoring,experiments show that the proposed scoring model had smaller variance and deviation rate,better stability and closer to the results of manual scoring.

关 键 词：相似度组合 Word2vec Doc2vec 文本相似度 自动评分 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]