关于滤除低质量的工人的相关文章：
JMLR-12 Raykar Eliminating Spammers and ranking annotations for crowdsourced labeling tasks
NIPS-14 Reputaiton-based Worker Filtering in Crowdsourcing
COLT-17 Efficient PAC Learning from crowd
众包中推导出 PAC 样本复杂度上界的文章：
IJCAI-16 Cost-saving effect of crowdsourcing learning
IJCAI-18
COLT-17
Arxiv-19
众包中对工人进行（隐式）聚类，一般都是假设一个生成式模型，然后依照此模型求解模型参数，工人就自动地聚到了一起。相关的 paper 有：
1.Community-Based Bayesian Aggregation Models for Crowdsourcing
2.Clustering Crowds （一个日本人的工作）
3. Distinguishing subjectiveness from difficulty
4. ICMl 18 Analysis of Minimax Error Rate for Crowdsourcing and Its Application to Worker Clustering Model
4. True Label + confusion matrix: 多个工人共享一个 confusion matrix