A Crowdsourcing Truth Inference Algorithm Based on Hypergraph Neural Networks
Zhaoan Dong, Yueyang Li, Lijun Gao, Zili Zhou · 2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · 2022
Crowdsourcing has become an economical and efficient way to obtain data, but the data obtained by crowdsourcing is often noisy. Due to concerns about human errors in crowdsourcing, it is necessary to infer the truth from the answers of multiple crowdsourcing workers. Traditional truth inference algorithms treat truth inference as a probability generation process and model the conditional dependency between variables. However, the fly in the ointment is that we need to make delicate assumptions about the priors of various variables for these methods. Additionally, the design of the generation process is also complex extraordinarily. With the popularity of graph neural networks, existing works try to exploit graph neural networks to solve the truth inference problem in crowdsourcing. Nevertheless, one edge in traditional graph structures can only connect two nodes which simply expresses the pairwise relationships, but ignores the high-order relationships among multiple nodes in the truth inference problem. Therefore, we propose a novel strategy which exploits the hypergraph neural network to solve the truth inference in crowdsourcing. We construct a hypergraph which contains task nodes, worker nodes, and candidate answer nodes and derive a new hypergraph neural network to learn the representations of nodes and the true labels. Besides, we exploit the latent correlation information between nodes of same types. In this paper, we explain our method through a case study.