在向量空間模式資訊擷取系統上，相關回饋是一種應用於提升擷取效率的技術。相關回饋的技術係在檢索過程中，由使用者對於系統檢索出的文件，進行相關或不相關的評估。過去的研究，主要在運用使用者回饋的資訊，修改使用者的興趣向量。本研究找出，過去研究在使用者相關或不相關的回饋文件中，未完全被研究過的資訊。這些資訊是關於字詞在相關或不相關文件中，所出現的各種狀態。本研究發展一個實驗性的資訊擷取系統與方法，以展示對於字詞出現狀態資訊的應用，並進行相關實驗，以研究這些方法是否具有效果。本研究實驗的結果顯示，字詞出現狀態的資訊是可以被抽取出來，並可應用於提升擷取效率。 Relevance feedback is one of the techniques applied in a vector-space-modeled Information Retrieval (IR) system to enhance retrieval effectiveness. The feedback process usually has the user rate the documents retrieved as relevant or non-relevant. Most past studies apply the information of document relevance to the modification of the vector that is used to manifest the user’s information interest. In this study, we have identified additional information obtained from relevance feedback that was not fully studied in the past from the rated relevant/non-relevant documents for application. The information pertains to is about the situations of term appearances in the relevant/non-relevant documents. We have developed a method together with an IR system to demonstrate the application of the information of term appearance situation. Experiments have also been conducted to study its effect. The experimental results preliminarily show that the information of the term appearance situation could be extracted and appropriately applied to enhance retrieval effectiveness.
