Abstract—Crowd-powered search is a form of crowdsourcing scheme which involves collaborations among voluntary Web users. Most popularly known episodes are succeed, while search tasks often failed in fact. In this research, we analyzed the factors which related to the performance of crowd-powered search though human flesh search (HFS) episodes, and predicted search performance based on these factors. We have analyzed 2.3 million microblogs about HFS which involved more than 1.3 million users over 2 years in Sina Weibo—the most popular social media site like twitter in China. Some useful features are found. Based on these features, we predict the performance of HFS episodes based on random forest method. The results of classification shown that our model performed good at differentiating these succeed and failed episodes automatically.

Index Terms—crowdsourcing, crowd-powered search, human flesh search, social computing, random forest, online collaboration

Cite: Tao Wang, Weiming Zhang, Cheng Zhu, Kaiming Xiao, Zhong Liu, and Baoxin Xiu, "Prediction of Crowd-Powered Search Performance Based on Random Forest," Journal of Industrial and Intelligent Information, Vol. 3, No. 4, pp. 293-298, December 2015. doi: 10.12720/jiii.3.4.293-298