摘要 本研究主要的目的為發展國小摘要寫作評量系統。評量的系統主要參考「潛在語意分析」的摘要寫作評量的方法,從教材及學生的摘要作品中,找出關鍵詞做為評量的依據。雖然在本研究的方法中,並未利用SVD(singular value decomposition)的技術來建立教材的潛在語意空間。但是經由關鍵詞的比對,仍然可以達到評量的效果。除了以常用的相關(correlation)函數來計算文件的相似程度外,在本研究中加上文件摘要效率的觀念及Entropy的觀念融入摘要評量的方法中。 本研究以台中縣富春國小六年級的79名學生為實驗的對象。在經過摘要寫作及心得寫作的實驗過程及單元成就測驗後,評估透過本評量系統所獲得的評量指標與兩位教師評量結果和學生成就測驗之間的相關性。研究結果有以下發現: (1)本評量系統以關鍵詞為摘要寫作評量的依據,能夠達到令人滿意的效果;(2)本研究將文件摘要效率的觀念及Entropy的觀念融入摘要評量的方法中,並經由實驗中證實其評量效果不僅可提高與外在效標之間的關聯效度,更可有效降低與字數間的相關程度;(3)以教師的摘要為標準來評量學生的摘要寫作能獲得最佳的評量效果,然而直接以教材內容為標準來評量學生的摘要寫作,也能獲得滿意的結果。 關鍵字:關鍵詞、摘要、寫作、摘要寫作評量
Abstract The main purpose of this research is to develop a summaries evaluation system. The summaries evaluation system is based on Latent Semantic Analysis method. But it does not utilize the Singular Value Decomposition technical method just because its time-consuming process to build the latent semantic space. In this research, we present several approaches to grade the student’s summaries writing. It utilizes the corresponding weights of the keywords for textbook and student’s summaries. And we tried to match the sixth-graders’ summaries against a set of six summaries evaluation approaches. The average correlation based on the grades of the two teachers with our approaches’ scores for the 79 of the students’ summaries was quite high. Therefore, these scores are quite comparable to how an experienced teacher rates these summaries. Thus, based on evidence from these trial, the summarization software could work well. A comparison of the average grades given by two teachers likewise showed no difference in quality of the students’ summaries related to condition. The agreement between the two teacher graders was very high also. This paper proposes several approaches to generate summaries score. The experimental results show that two or more of our approach can easily perform better than the traditional cosine approach. Among of them the cos2 and compromise methods perform better than others. Keywords: keyword、summaries、summaries evaluation