基於Google 搜尋趨勢之景氣對策信號預測

景氣對策信號是我國為了衡量經濟景氣概況的重要指標，是由一些能夠反映景氣變化的總體經濟變數以統計方式處理，編製而成。由於這些經濟變數必須由政府各個部門彙總後來取得，各部門的溝通和數據的處理需要耗費一段時間，導致景氣對策信號的發布延遲，無法及時反映當前的景氣狀態，政府與企業無法及時根據景氣信號制定相關策略。因此，為了縮短上述的落差，本研究運用人民在網路上有關於景氣的搜尋趨勢取得代表景氣的關鍵字詞，再取得這些關鍵字詞的Google搜尋趨勢，將這些資料利用機器排序學習，訓練出可以預測景氣對策信號的模型。研究結果顯示該模型對於預測未來景氣對策信號的狀態，與相關研究相比有較高的準確度。由於資料的來源為網路上的搜尋日誌，相較於傳統經濟變量更具有即時性和好取得的特性，能夠降低目前景氣狀態發布的延遲。

關鍵字

Google搜尋趨勢；特徵選取；機器排序學習；預測景氣對策信號；關鍵字搜尋

並列摘要

Prosperity countermeasure signal is an important indicator to measure the general situation of economic prosperity in Taiwan. The signal is compiled from some overall economic variables that can reflect changes in the economic climate. These economic variables are aggregated and obtained by various government departments. The communication and data processing of various departments takes a period of time, resulting in a late release of the economic countermeasure signal, which cannot reflect the current economic status in a timely manner. Therefore, the government and enterprises cannot formulate relevant strategies in time according to the economic signals. In order to solve the problem of the delay of economic signals, this research uses people's search trends about the prosperity on the Internet to reflect the current situation of the prosperity, and then uses machine learning to train a model that can predict the economic countermeasure signals. The research results showed that the model has higher accuracy in predicting the state of future economic countermeasure signals compared with related studies. Since the source of the data is the search log on the Internet, it is more timely and easy to obtain compared with traditional economic variables, which can reduce the delay in publishing the current economic status.