隨著網際網路的發展與社群網站的興起，使用者產製的內容日趨增多，對於發生事件的討論也越來越多，更有使用者在社群媒體上揭發不為人知的事件與行為，形成俗稱的「爆料」，如2016年的復興航空停飛事件。若是想瞭解在社群網站上之討論事件的來龍去脈，使用者需逐一檢視大量的貼文、搜尋相關資訊、並且自行整理事件的過程，較為耗費時間與人力成本。鑑於此，本研究希望提出一套在社群媒體中進行事件偵測與追蹤的流程，降低使用者在搜尋與整理事件的負擔。 本研究進行相關文獻探討，整理其中的問題後，先研擬事件偵測與追蹤的演算法，並依此建構雛形系統，將文章以事件進行群聚與排序後呈現，再設計實驗以驗證雛形系統之準確率與使用者的滿意度。實驗結果發現約有78％的受測者給予系統4分以上的滿意度，改進建議包含希望議題多元化與系統個人化；而F-measure為0.64，推測可能因事件本身的討論發散程度，而影響系統分析的準確度，最後提出系統貢獻與可改進之處，為後續研究的參考。 As more and more people use social networking website, the user-generated content is increasing, and if somebody wants to know the context of the discussion on the community website, users need to review a large number of posts and self-collate events. Thus, the research objective is developing a system of event detection and tracking in the social media, reducing the burden on users to understand the events. In this study, the relevant literatures are discussed and the algorithm of event detection and tracking is developed after finishing the problem. The prototype system is constructed and the experiment is designed to verify the system performance and user satisfaction. The results of the experiment show that about 78% of the subjects to give the system more than 4 points of satisfaction, improve the proposal to enhance the issue about issue diversification and system personalization, and the average F-measure is 0.64. Finally, make research contributions and future development for the follow-up study reference.
