隨著網路的普及，人們在網路上分享影片並觀看視頻，已成為人們每天都在做的事情，然而，如何在眾多影片中找到熱門的影片，成為影片管理者、廣告商、影片製造商所關心的事。過去研究在預測熱門影片，依賴過去的歷史資料來作預測，當無歷史資料時，將會面臨傳統推薦上的冷啟動問題。本研究以剛上傳的影片作為預測，找出三十六個預測變數建模，並以五個機器學習的分類演算法(類神經、貝式、支持向量機、羅吉斯回歸、決策樹)作為集成學習ENSEMBLE方法，來建立一至十週的預測模型。本研究另外針對各個屬性構面作為挑選，並探討影響影片熱門度的關鍵屬性。本研究結果顯示，本研究的預測模型，均有很好的預測能力，並且可以解決新影片上傳的冷啟動問題。;Predicting video popularity is an important task involved in managing video-sharing sites. Although many previous studies have investigated this problem, a weakness common to these studies is that their predictions rely on video access data from the past. In other words, they cannot predict the popularity of newly uploaded videos. To handle this cold start problem, this study focused on building prediction models that use only the data available at the time when a video is initially uploaded. Through supervised learning methods, this study employed prediction models to predict the popularity of videos. To further improve the overall accuracy of the prediction, we used an ensemble model to integrate these classification results to obtain the most accurate prediction. The empirical evaluation indicated that the models are effective for predicting the popularity of a video and that our model can solve the cold start problem of video popularity prediction.
