Abstract: With the advent of the era of big data, the association of cross-platform virtual accounts has become an urgent problem in the field of network supervision. This paper took user text data such as Weibo and WeChat as the research object. Through sampling and manual labeling of the data, a quantitative analysis of the natural labeling behavior of cross-platform accounts in the open social platform was carried out. And from this, a cross-platform virtual account association method based on the users natural annotation was proposed. Aiming at the characteristics of natural labeling, this method constructed a virtual account recognition model based on the characteristics of upper and lower words, used a two-category deep learning model for the same authentication of nicknames and users, and realized the recognition of cross-platform virtual accounts with the recognition accuracy rate reaching more than 85%.