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

Abstract: 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’ know-ledge 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.

Key words: Automated subjective question scoring,Latent semantic analysis (LSA),Latent Dirichlet allocation (LDA),Absolute accuracy rate,Adjacent accuracy rate