協同過濾推薦最佳配對於人力運算效能改進之研究

Abstract

隨著網際網路的普及，人力協同運算（human computation）應運而生。然而全 球20億的網路使用者，各自具備不同的背景能力； 以目前人力協同運算系統的隨 機分工方式，使用者不易達成共識， 也難以針對其擅長的領域做貢獻，降低系統 的效率。 為了解決這個問題，我們利用推薦系統（recommender system）中的協同過濾 方法（collaborative filtering）， 針對使用者的參與記錄，匹配適當的同伴 與其擅長領域的問題。我們以Luis von Ahn教授設計的ESP Game為本，在Amazon Mechanical Turk人力平台上邀集910人參與共4757場遊戲。實驗結果驗證藉由推 薦系統提供使用者最佳配對，能提高人力協同運算的效率。

With the popularity of the Internet, humancomputation came into being. However, there are 20 million Internet users from the whole world, each with a different background skills. Current human computation system with random division of labor making users hard to reach a consensus with each other and contribute on their expert domains reduces the efàciency of the system. To solve this problem, we use the collaborative f iltering approach of recommendersystem to match the appropriate partner and the problem in their expert domain with user history. Based on professor Luis von Ahn’s ESPGame, we design the experiment on the crowdsourcing platform – AmazonMechanicalTurk. And total 910 users involve and 4757 games are record. Experimental results demonstrate that the recommender system provide users the perfect match and improve the efàciency of human computation.