發明
美國
12/631,103
US 8,140,543 B2
時序性文字摘要與故事情節演變分析演算法METHOD OF TOPIC SUMMARIZATION AND CONTENT ANATOMY
國立臺灣大學
2012/03/20
一種時序性文字摘要與故事情節演變分析演算法,係將一事件之相關文章拆成一群彼此不重疊之內容區塊(Blocks),並作為事件摘要之基本單位。基本上係以一限制最佳化法(Constraint Optimization Methods)描述故事主軸(Themes)辨識問題,並以矩陣(Matrix)BTB之特徵向量(Eigenvector)表示內容獨立之故事主軸,以從該事件中找出多個故事主軸;接著分析特徵向量內之數值變化以找出各故事主軸之重要發展且產生次事件摘要;最後計算時序相似度而連接各內容相似之次事件以構成事件之故事演變圖。藉此,能產生品質較佳之時序性新聞摘要,且經仔細分析本發明所產生之故事演變圖後,能證明其係可確實地標示出事件內重要之發展,且亦能有效地陳述各發展之因果關係,以達到減少使用者了解一熱門事件之閱讀負擔。 The main purpose of the present invention is to analyze documents related to a topic through an eigenvector-based algorithm for generating a summary and an evolution graph of the topic. The second purpose of the present invention is to obtain a temporal topic summary having a good quality with a consideration of topic temporality. The third purpose of the present invention is to faster select representative sentences, paragraphs or documents for a topic while a compression ratio of summary is higher. The third purpose of the present invention is to obtain an evolution graph showing important events in the topic and indicating cause-result relationships between the events for reducing difficulty in understanding an evolution of the topic.
產學合作總中心
33669945
版權所有 © 國家科學及技術委員會 National Science and Technology Council All Rights Reserved.
建議使用IE 11或以上版本瀏覽器,最佳瀏覽解析度為1024x768以上|政府網站資料開放宣告
主辦單位:國家科學及技術委員會 執行單位:台灣經濟研究院 網站維護:台灣經濟研究院