A method of managing network usage by defining a set of linguistic patterns, where each linguistic pattern is associated with a condition that is to be monitored. Network packets are captured during transmission and analyzed to identify linguistic patterns. Captured network packets are scored based on similarity of at least one linguistic pattern to one or more of the defined set of linguistic patterns. When a packet that is scored above a specified threshold value is identified, at least one responsive action is implemented. In this manner, a system implementing the method is able to identify network traffic that is associated with prospective malicious activity and thereby provide an early warning before damage has occurred.