| 摘要: | 隨著資訊科技和網路的發展，資訊傳播的速度比以往更快且更便利。但人們在傳播資訊時，通常不驗證此資訊的來源和可信度，尤其是在社群媒體平台。如此未經驗證的資訊在網路上流竄稱之為網路謠言(Online rumor)。而現今網路謠言的氾濫，不僅引起社會恐慌，還改變與論方向。為了增加人們對於謠言的認知，在實務上，已有民間組織建立謠言查詢網站，例如Cofacts、Mygopen和蘭姆酒吐司。但這些網站是依靠人工檢查的方式來變是網路謠言，需要大量的人力進行查驗；在學術上，也有許多研究提出深度學習和機器學習的方法，但深度學習的方法若是模型架構過大，則會導致訓練模型的過程耗費時間。而機器學習的方法雖有不錯的準確率，但無法解決語意的問題。此外，若使用者對於模型預測後的結果是有疑慮的，則需要有一套機制能經由過往案例的參考，以推論的方式辨識網路謠言。
因此，本研究應用本體論和機器學習模型，使模型預測的過程能處理語句上反義的關係。此外，本研究結合了基於案例推論，使用者若對預測的結果是有疑慮時，能以半自動的方式進行案例推論，達到網路謠言辨識。結合上述兩點，本研究實做了一套網路謠言辨識系統－MOCIR(Machine learning Ontology Case based reasoning Identify Rumor)，以Web-based和Linebot作呈現。而在最後比較既有的機器學習模型和實務界以繁體中文為主的系統之方式來驗證本系統。
;With the development of information technology and the Internet, the speed of information spreading has significantly increased, people on social media platforms usually are not able to effectively verify the source and credibility of the information. Unverified information spreading on the Internet was called online rumor. The rumor has become a severe problem, not only caused the social panic, but also changed the direction of public opinion. To increase people′s awareness of rumors, non-governmental organizations have established rumor query websites, such as Cofacts and Mygopen, which rely on manual verification methods on identifying online rumor. In academia, There are many researchers proposed deep learning and machine learning techniques for identifying rumor. However, if the architecture of deep learning model is too large, the process of training would be time-consuming. Although the machine learning model has an excellent accuracy, but it can not solve the sematic problem. In addition, if the user is unacceptable about the prediction results by the model, then a mechanism is needed to identify the online rumors by reasoning method and referring to similar cases.
Therefore, this research applies machine learning techniques and ontology models to predict online rumor and deal with antisense problem. Moreover, if the users do not accept the predicted results, then they could use case-based reasoning in a semi-automatic way to achieve online rumor identification. In conclusion, this research has implemented the proposed methodology into an online rumor identification system, and the users could access our system by website or Linebot. The system was verified by comparing the related machine model and the traditional Chinese-based system in practice. |
