封閉與開放式群眾智慧比較: 以傳染病資訊市場為例

Abstract

在本篇論文中，我們以台灣的三個不同傳染性疾病案例比較了開放式群眾智慧以及封閉式群眾智慧的預測結果表現，預測時間區段為2010年三月至2010年十月。我們利用一個由公共衛生專家及醫療人員所共同參與的封閉性資訊市場做為封閉式群眾智慧的代表。另一方面，我們挑出了一些與疾病有關係的關鍵字，並從Google找出這些關鍵字過往的使用者搜尋次數，再用這些搜尋次數紀錄與從台灣疾病管制局所得到的疾病歷史紀錄建立線性迴歸模型，最後用這個線性迴歸模型所做出來的預測結果來代表開放式群眾智慧的預測結果。我們也另考慮了「時間延遲」效應及「關鍵字數」效應來改善代表開放式群眾智慧的預測結果。除此兩種方法之外，我們還使用了歷史平均法做為我們進行比較時的基準。 我們用以下三個不同的傳染性疾病案例來討論我們的比較結果: 1.) 流感重症病患病例數, 2.) 腸病毒病患比率, 及3.) 類流感病患比率。在這三個例子中，我們主要使用平均誤差絕對值做為我們的比較指標，另外相關係數的結果也併呈於結果的討論中。我們依不同的案例去比較這幾種方法的預測結果，並討論它們的特性及影響預測結果的因素。整體的結果顯示封閉式群眾智慧的表現為最好，開放式群眾智慧的表現為次佳，而歷史平均法的表現則為最差。

In this thesis, we compared the prediction performance of opened wisdom of crowds and closed wisdom of crowds in three different disease activities in Taiwan from March 2010 to October 2010. We used the prediction results coming from a group of healthcare workers and public health professionals participating in the closed prediction market to represent the closed wisdom of crowds. In the other hand, we fit the past search query frequency data of terms that related to the disease from Google with the historic disease activity data from Centers for Disease Control (CDC) in Taiwan to train the linear regression models, and use these linear regression models to generate the prediction results that represents the opened public wisdom of crowds. We also tuned the “time lag” and “term number” to improve prediction of opened wisdom of crowds. In addition to these two approaches, we incorporated the historic average as our comparison baseline. The prediction results are been compared in three different cases: 1.) patient number of severe complicated influenza (流感重症病患病例數), 2.) proportion of enterovirus patients (腸病毒病患比率), and 3.) proportion of parainfluenza patients (類流感病患比率). In each case, we used the average absolute error as the main metric to compare their prediction performances, while the correlations of coefficients are discussed. The results are shown in each case. The overall results showed that closed wisdom of crowds performs the best among three methods, the opened wisdom of crowds is the second, and the historic average is the worst.