Methods and systems are presented for analyzing transactions conducted through user accounts with an online service provider based on graph analysis. A graph is generated based on a set of seed accounts that are determined to be involved in suspicious activities. The graph includes a set of seed nodes representing the seed accounts, and a set of nodes representing user accounts that are connected to the set of seed accounts in downstream transactions. A random walk traversal based on multiple dimensions is performed on the graph to determine nodes that are closely related to the set of seed nodes. Transactions conducted through the seed accounts and accounts corresponding to nodes that are closely related to the set of seed nodes are analyzed to detect any malicious activities. The graph may also be presented according to a layered hierarchical structure for better visualization of transaction flows through the accounts.