Behavior-based community detection in enterprise information networks
Methods and systems for detecting host community include modeling a target host's behavior based on historical events recorded at the target host. One or more original peer hosts having behavior similar to the target host's behavior are found by determining a distance in a latent space that embeds the historical events between events of the target host and events of the one or more original peer hosts. A security management action is performed based on behavior of the target host and the determined one or more original peer hosts.
1. A method for detecting host community, comprising:
embedding historical events, recorded at a target host, in a latent space using a negative sampling that approximates a maximized conditional probability that an event will occur at a target host;
modeling the target host's behavior based on the embedded historical events, wherein the target host is modeled as a context of process-level events and as a context of all hosts reached by network events of the target host and wherein each process-level event on the target host is modeled as a conditional probability of the target host given the event;
determining one or more original peer hosts having behavior similar to the target host's behavior by determining a distance in a latent space that embeds the historical events between events of the target host and events of the one or more original peer hosts; and
performing a security management action based on behavior of the target host and the determined one or more original peer hosts.
2. The method of claim 1 , wherein determining one or more peer hosts comprises clustering hosts based on distance in the latent space.
3. The method of claim 2 , wherein clustering comprises identifying a set of initial cluster centroids and iteratively updating the centroids after assigning hosts to a closest cluster.
4. The method of claim 1 , wherein performing the security management action comprises determining an anomaly score based on a comparison between behavior of one or more new peer hosts and behavior of the one or more original peer hosts.
5. The method of claim 1 , wherein performing the security management action comprises determining an anomaly score based on a similarity between new events recorded at the target host and new events recorded at the one or more original peer hosts.
6. The method of claim 1 , wherein performing the security action further comprises automatically performing at least one security action selected from the group consisting of shutting down devices, stopping or restricting certain types of network communication, raising alerts to system administrators, and changing a security policy level.
7. The method of claim 1 , wherein each network event is modeled as a conditional probability that one host issues a network event that connects to another host.
8. A system for detecting host community, comprising:
a host behavior module configured to embed historical events, recorded at a target host, in a latent space using a negative sampling that approximates a maximized conditional probability that an event will occur at the target host, and to model the target host's behavior based on the embedded historical events, to model the target host as a context of process-level events and as a context of all hosts reached by network events of the target host, and to model each process-level event on the target host as a conditional probability of the target host given the event;
a peer host module comprising a processor configured to determine one or more original peer hosts having behavior similar to the target host's behavior by determining a distance in a latent space that embeds the historical events between events of the target host and events of the one or more original peer hosts; and
a security module configured to perform a security management action based on behavior of the target host and the determined one or more original peer hosts.
9. The system of claim 8 , wherein the peer host module is further configured to cluster hosts based on distance in the latent space.
10. The system of claim 9 , wherein the peer host module is further configured to identify a set of initial cluster centroids and iteratively update the centroids after assigning hosts to a closest cluster.
11. The system of claim 8 , further comprising an anomaly score module configured to determine an anomaly score for use by the security module based on a comparison between behavior of one or more new peer hosts and behavior of the one or more original peer hosts.
12. The system of claim 8 , further comprising an anomaly score module configured to determine an anomaly score for use by the security module based on a similarity between new events recorded at the target host and new events recorded at the one or more original peer hosts.
13. The system of claim 8 , wherein the security module is further configured to automatically perform at least one security action selected from the group consisting of shutting down devices, stopping or restricting certain types of network communication, raising alerts to system administrators, and changing a security policy level.
14. The system of claim 8 , wherein the host behavior module is further configured to model each network event as a conditional probability that one host issues a network event that connects to another host.