Evaluating the importance of terms with Latent Semantic Analysis: An application study

• Main Authors: Pai, Kai-Chih, 白鎧誌
• Other Authors: Kuo, Bor-Chen
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/00960434602767765989

碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 99 === The purposes of the present study was to develope the indicators of the significance of terms and to examine the effectiveness of these indicators by comparing it with human scoring. LSA-based automated scoring system for word construction and sentence construction tasks were developed by employing the indicators. The utility of the automated scoring system was examined by comparing with the results of human scoring. Also, the relation of word and sentence construction and, reading comprehension was investigated. The results of the study showed: 1. The result of the significance of terms was comparable between the indicators and human scoring. The term vector length and entroy were moderately correlated with human scoring, whereas the indicators of the significance of terms correlated highly with human scoring in the sentence construction task. 2. The LSA-based automated scoring for sentence construction and the human scoring were highly correlated, which implies that LSA-based automated scoring is effective. 3. Word and sentence construction tasks predicted reading comprehension in both means of scoring, which shows that LSA-based automated scoring works as well as human scoring 4. Comparing the results of traditional scoring and LSA-based automated scoring, LSA-based automated scoring explained more variance in reading comprehension.