在網路的發達的時代，新聞資訊不斷地在網路世界中擴張散播，過多的新聞使我們 需花費時間閱讀全文才能找到想找的資訊，因此本研究提出一應用語句字詞關係於多文 件自動摘要之方法，能自動找出文件中的重點做為摘要，如此即可讓讀者節省閱讀全文 的時間，本研究將文件中每一語句視為一筆交易資料，並使用關聯規則演算法挖掘出頻 繁項目集，利用頻繁項目集計算產生關聯字詞，最後依照語句所含之關聯字詞，擷取最 高語句計分之語句產生摘要，提升從多文件擷取最佳語句作為摘要的準確率。本研究使 用DUC 2004 新聞文件集進行DUC 2004 task2 之實驗，作出665 bytes 之摘要，經過 ROUGE 評估摘要品質，本研究所提之方法能有改善多文件自動摘要之潛能。;With the quick development of the internet, the news spread worldwide in minutes, the presence of too much information make us hard to understand the issue and spend too much time on reading the news to get what we want. Therefore, in this research, we aim to produce an extract-based summary to provide readers a quick review of the news. In the research, we attempt to use association rule to extract the relevance terms of sentences and apply it on documents summarization. In the experiments, the results show that applying relevance terms of sentences on multiple documents summarization could be effective in improving the precision of summarization.
