微網誌是一個新興的社群互動平台,能立即反應使用者當時的心情,具有即時性,但因為其內容長度限制,通常會在短時間內產生大量簡短的訊息。這樣的微網誌社群具有兩項問題:第一,大量訊息之間會有各自的討論主題,並非所有使用者都有興趣,這讓使用者難以閱讀自己真正有興趣的訊息。第二,其內容字數的限制,使每則訊息涵蓋的資訊過於稀疏,這讓使用者難以分辨訊息的重要性和可靠性。 為了解決上述問題,本論文導入影響力分析和意見分析探討微網誌社群,我們提出一影響力模型,以回應、轉載等社群結構關係,分析訊息的影響力。其次我們提出一意見模型,根據訊息的內容,偵測熱門主題與訊息,以分析訊息回應的意見傾向。 實驗以Plurk的中文訊息為主,藉由文件分類與實際案例來觀察所提方法之效果。實驗結果顯示影響力分析可以有效分類出流行訊息,其最佳的F-measure為86%;而意見分析可以有效偵測使用者對於熱門事件的喜好程度,以2012台灣總統選舉支持率預測為例,其預測的平均絕對誤差為1.26%,準確度為98.74%,證實本論文提出方法可以有效分析微網誌訊息。
With the popularity of microblog as a new social communication platform, users can easily share their feelings and opinions within 140 characters, which have attracted many research efforts. There are some issues for microblogs. First, users tend to write many short messages in different topics, which make other users difficult to find topics that really interest them. Second, each message is limited to 140 characters, so users are difficult to get vital and reliable messages that they need. In this paper, we designed two novel models on Plurk: influence analysis model, and opinion analysis model. First, influence score is calculated from the number of replurks, responses, and likes. Then, hot topics are identified from popular messages and net opinions from responses are accumulated as the overall rating. Our experimental results show a high F-measure of 86%, for classifying popular discussions with influence score. In the case of the Taiwan presidential election forecast, the MAE of prediction is 1.26%, which shows the effectiveness of the proposed approach.