The present invention relates to a frequent co-occurrence account mining method in a transient online transaction, to effectively solve a problem of implementing cross-data stream tracking and auditing for an online behavior of a user by expanding behavior auditing of a single online account to multiple relevant applications and accounts. The method comprises: based on a network distribution technology, at a network port mirror network raw data, based on fluctuation keyword matching and regular expression filtering, identifying data streams corresponding to different applications in a network, extracting user online data on a social application, collecting and sorting according to time and an online address, to form an online transient transaction database, searching transaction data in a transaction database corresponding to a to-be-analyzed user online raw stream for frequent item sets, calculating an overlapping rate between the item sets, and combining the frequent item sets to obtain a plurality of network accounts belonging to one social person. The method provided by the present invention is simple in operation, and relevance between network accounts that hide in a network communication stream can be found accurately.