TED Talks巡覽圖之建構-最佳化方法之應用

Abstract

TED Talks 網站擁有大量且豐富的影片資源可供大眾學習科技、娛樂、以及設計等相關知識。然而，現今之TED Talks影片搜尋系統並無法藉由便利的視覺呈現方式協助使用者從大量的搜尋結果中迅速找到標的影片。本文提出一種直覺、易於理解的TED Talks影片巡覽圖(Navigation-Graph)，這種視覺化方法可根據影片與關鍵字之間的關係，採用最佳化模型產出TED Talks影片及關鍵字於該巡覽圖面上之最佳擺放位置，並強化相關的視覺參數，以協助使用者可以便捷、迅速地尋得欲觀看之相關影片。為了改善巡覽圖中影片與關鍵字配置模型的效率，本文另提出一強化型TED Talks影片巡覽圖(Enhanced Navigation-Graph)，當使用者採用關鍵字搜尋出大量的影片時，本方法首先會根據影片內關鍵字之間的關係，將關鍵字適當地安排於巡覽圖的外環，隨即依據影片與關鍵字的關係將相關影片放置於巡覽圖的內部。因此，使用者能夠依其直覺迅速地參照視覺化呈現的線索，並採用本文所提出之4S(Strengthening-Seeing-Sailing-Selecting)搜尋策略於大量影片中篩選有興趣的主題影片，或以視覺線索規劃其個人化的影片瀏覽路徑。實驗評估結果顯示，所提出之方法可大量地降低計算複雜度、改善系統運算效能；而且透過此TED Talks影片巡覽圖亦能夠有效且快速地協助使用者找尋他們偏好的影片。

TED Talks website is currently one of the most well-known systems providing technology, entertainment and design videos. However, the current TED Talks system is not convenient for users to find their target videos from a large set. This study proposes a Navigation-Graph method to visualize the relationships among numerous tags and videos, thus to assist a user to navigate on the Navigation-Graph to reach their videos. Besides, we modified Navigation-Graph as an Enhanced Navigation-Graph to improve the computational time of problem solving for allocating videos and tags. Given a large set of videos with associated tags, the proposed method first allocates all selected tags on the ring of the Enhanced Navigation-Graph, then to position all related videos on the inner of the Enhanced Navigation-Graph using optimization models. After that, users can efficiently find their preferred videos via 4S (Strengthening-Seeing-Sailing-Selecting) navigation strategies. Experiment results explicate that the proposed method performs better than the existing navigation method and current official TED Talks displaying system for helping users find their desirable videos.