基于改进的LDA模型的中文主观题自动评分研究  

Automated Scoring Chinese Subjective Responses Based on Improved-LDA

作　　者：罗海蛟[1] 柯晓华 

机构地区：[1]广东外语外贸大学思科信息学院,广州510006 [2]广东外语外贸大学语言工程与计算实验室,广州510006

出　　处：《计算机科学》2017年第B11期102-105,128,共5页Computer Science

基　　金：广东省自然科学基金资助项目(2015A030313575)资助

摘　　要：主观题自动评分(Automated Scoring Subjective Responses,ASSR)在语言学习与语言测试领域的诊断信息及信度方面具有重要的应用前景。将主题模型中的隐含狄利克雷分配(Latent Dirichlet Allocation,LDA)引入到中文主观题自动评分中,提出了一种结合专家知识的改进的LDA模型,并采用了一种综合文档-隐含主题概率向量及隐含主题-核心词项概率向量的文本特征表示方式。实验对比了改进的LDA与潜在语义分析(Latent Semantic Analysis,LSA)的自动评分效果,结果表明改进的LDA模型在中文主观题自动评分中切实有效。Automated scoring subjective responses (ASSR) have great promise for providing diagnostic information and reliability to aid language learning and testing.In the presnt study,we introduced the latent Dirichlet allocation (LDA) into an automated scoring task with Chinese subjective responses,and an improved LDA model with experts' knowledge was proposed.In the novel model,we proposed a text feature representation approach integrating document-latent topic probability vector and latent topic-core terms probability vector.Experiment results show that the improved-LDA is better than LSA in terms of the autoscoring performances.The findings of this study highlight the model selection in application of automated scoring Chinese responses with language testing.

关 键 词：主观题自动评分 潜在语义分析(LSA) 隐含狄利克雷分配(LDA) 绝对一致性 相邻一致性 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]