近年來網路的發達，讓資訊的傳播更為快速，人們也能隨時接收到許多新的資訊，其中像是新聞訊息更新非常迅速，但是過多、更新過快的新聞內容讓讀者需要花費更多的時間去閱讀每篇新聞文章的完整內容，以掌握新聞的重點。因此本研究的目的在於提出一個應用語句中字詞關聯圖形網路於多文件摘要的方法，找出文章中的重點摘要，讓讀者可以花較少的時間了解新聞的內容。在文件中經常一起出現的字詞組合可能含有其資訊，本研究以關聯規則找出語句中經常一起出現之字詞做為字詞關聯項目，並用其作為節點建立圖形網路，利用圖形中心性找出圖形中較重要之節點，計算語句所涵蓋之關聯規則項目計算語句分數，再根據語句權重分數挑選最高分的語句做為摘要。本研究使用DUC 2004新聞集並進行Task2實驗，輸出665bytes之摘要，透過ROUGE及專家摘要來評估摘要品質。;Internet develops quickly and makes information spread worldwide. However, update of information in minutes makes people spend much time to read news. Therefore, the purpose of this research is to generate an extractive-based summary for people to have a concept of news. We attempt to apply association rule for extracting relevance terms of sentences from documents and use a graph-based method for calculating the scores of relevance terms and sentences, and then we select the sentence which has higher score to produce summarization of multi-documents. The results of our experiments show that the ROUGE value of applying relevance terms on graph-based multiple documents summarization method could be effective in summarization.
