Crowdsourcing innovation, as a new innovation pattern, helps companies reduce the risks and costs of innovation, which has received widespread attention and practical application. What is critical for improving crowdsourcing innovation performance is to understand the heterogeneity of participating users deeply, guide and motivate users to participate actively. Based on the typical characteristics of crowdsourcing innovation communities, this paper proposes a model integrating social network analysis (SNA) & K-means clustering algorithm to identify participants' roles and conducting empirical research with Xiaomi MIUI community. The result indicates that users can be divided into nine categories: active user, positive user, negative user, bystander, creative contributor, faithful supporter, tourist, and new participant. In order to provide decision support for enterprises to govern crowdsourcing innovation virtual community effectively and improve innovation performance, this paper analyses the behavioural characteristics of each user role from two dimensions: interaction behaviour and contribution behaviour.