Aggregating Crowd Wisdom with Instance Grouping Methods

摘要：

　　With the blooming of crowdsourcing platforms,utilizing crowd wisdom becomes popular.Label aggregation is one of the key topics in crowdsourcing research.The goal is to infer true labels from multiple labels provided by different users.Most researchers make their efforts in modeling user ability and instance difficulty.However,these methods may suffer from sparsity of labels in practice.In this paper,we consider label aggregation from the view of grouping instances.We assume instances are sampled from latent groups and instances in the same group share the same true label.A probabilistic graphical model named InGroup (Instance Grouping model) is constructed to infer latent group assignments as well as true labels.Further,we combine user ability and group difficulty into InGroup to achieve a better model called InGroup+ (InGroup Plus).The experiments conducted on a real-world dataset show the advantages of instance grouping methods compared with other methods.

关键词： Crowd wisdom Label aggregation Instance grouping Probabilistic graphical model

作者: Li’ang Yin Zhengbo Li Jianhua Han Yong Yu

作者单位: Computer Science Department,Shanghai Jiao Tong University,No.800 Dongchuan Road,Shanghai,China

会议类型: 国际会议

会议名称: International Asia-Pacific Web Conference(第18届国际亚太互联网大会)

会议地点: 苏州

会议语种:英文

页码: 468-479

在线出版日期: 2016-09-23（万方平台首次上网日期，不代表论文的发表时间）