潛在語意分析之摘要評量系統─探討特定領域文章、關鍵詞、奇異值保留個數對建構語意空間及評量效度之影響

Abstract

本研究主要目的為探討「特定領域文章 (P)」、「關鍵詞 (K)」及「奇異值保留個數 (S)」三項因素對建構語意空間及評量效度之影響。利用自行開發之系統建構由上述三項因素所組成之語意空間，並將其評量結果與教師人工評量結果進行相關性分析，以探究三項因素的最佳建議值。 本研究以二重國小五年級的 107 名學生為實驗對象。在經過兩篇摘要寫作測驗及相關性分析後，得到以下結論：兩個實驗之分析結果呈現 LSA 的評分在端午節文章中與兩位教師人工評量相關性皆達顯著水準；但在丹頂鶴文章中則呈現不一致現象。因此，以 LSA 作為摘要評量分數的效度在不同文體未有一致結論。The main purpose of this research is to observe how the three factors–specific domain article (P), the keyword (K) and the number of dimension reduction (S)–influence semantic space building and evaluation effects. Using the summarization evaluation system developed by this research to build the semantic space consist of three factors above. Then we analyze the correlation between cosine value cacluated by each semantic space and two teachers’s scoring. The participants are the 107 students of Er-Chong elementary school. After analysis of two summarizing experiments, the results of the research are: In experiment I, the correlation between two teachers’ grades and the LSA cosine is significant. But in experiment II, the correlation between two teachers’ grades and the LSA cosine is inconsistent. Therefore, base on the LSA cosine, we could not get the consistent results in different literary forms.