COVID-19仍持續威脅著世界各國的公共衛生，而有效地預測COVID-19確診以及死亡人數上升或下降的趨勢，將有助於研究人員和政策制定者通過將 COVID-19 推向正確的方向來降低死亡率和確診率，目前對於COVID-19疫預測皆為使用結構化資料進行預測，並沒有學者使用非結構化的資料進行預測。然而，在非結構化預測上，因為社群媒體的蓬勃發展，透過社群媒體的文本進行預測，在各領域上皆有許多學者以此進行實驗，因此本研究想要透過社群媒體上有關於COVID-19的文本進行疫情趨勢預測。 本研究主要是利用不同的情感分析方法，將社群媒體的文本產生每日情感分數，再結合結構化資料進行疫情預測，預測目標為確診人數變化以及死亡人數變化。本研究選用不同的情感分析方法(辭典法、情感分析套件、靜動態詞嵌入方法)，並使用三種不同的分類器，SVM、LSTM、Bi-GRU進行分類，去分析最為有效預測疫情趨勢的組合，最終，本實驗發現以動態詞嵌入方法RoBERTa搭配Bi-GRU有最佳疫情趨勢預測，在預測確診人數，其評估指標Precision最高達75.89%。 ;COVID-19 is continuing to threaten the public hygiene of countries around the world. An efficiently way to predict the trend of COVID-19 epidemic will help researchers and policy maker make the right decision to reduce the mortality rate and confrimed case rate.At present,All research on COVID-19 epidemic prediction is based on technical data, However, With the development of social media,Using social media texts to predict is common in various fields.Therfore, This research is mainly discussed about using different sentiment analysis methods to generate daily sentiment scores from social media texts,and combine technical data for epidemic prediction. This research selects different sentiment analysis methods(dictionary method, API, and dynamic word embedding sentiment analysis method),and uses three different classifiers ,SVM、LSTM、Bi-GRU for epidemic prediction. At the end of the research, we found that the dynamic word embedding sentiment analysis method RoBERTa with the epidemic prediction classifier Bi-GRU can predict the trend of COVID-19 epidemic with best combination. In predicting the number of confirmed cases, evaluation indicator precision is rasie to 75.89%.
