以使用者偏好為基礎之數位文件排版轉換系統

Abstract

在本論文中，我們提出了一套數位文件排版轉換系統。主要目的就是減低使用者在轉換數位文件排版時所需的時間與精力。我們探討了兩種文件被轉換的格式，投影片以及網頁。使用者可以藉由我們的系統，將一份PDF格式的論文，輕易的轉換為PPT格式的投影片。其中有關語意的部分，我們採用了一個簡易的標記工具，讓使用者在不影響他們的閱讀習慣之下，選擇他們真正需要的內容。在收集到資訊之後，我們將投影片中的圖文排版模擬成一個最佳化問題，而使用者偏好則被視做最佳化的限制。除此之外，我們也為網頁設計了排版轉換系統。在處理網頁的圖文排版問題時，則是利用了基因演算法來解決較為複雜的最佳化問題。相較於傳統投影片以及網頁費時費力的產生過程，我們透過了使用者的主觀測試，本系統在使用者花費的時間精力以及被轉換數位文件的品質上取得了不錯的平衡。

In this thesis, we proposed a user preference guided document layout adaptor to reduce the overhead of end users. We focus on two types of publishing media, presentation slides and webpage. In slides generation, semantic-level summarization is fulfilled by introducing an easy-to-use annotation interface, rather than being trapped in the difficult text understanding problem. Furthermore, we model media-specific principles of document layout design, together with user preferences, as constraints of optimization problems and implement an efficient layout optimization module. Users of the proposed slides generator can conveniently read/annotate a digital document and then obtain high-quality presentation slides on the fly. Besides, a webpage generator is proposed to further support the feasibility of document adaptation system. We accomplish the webpage layout optimization problem by genetic algorithm. Extensive experiments including subjective tests are performed to prove that the proposed system does make good trade-offs between user overhead and presentation quality.