Multi-signal analysis for compromised scope identification

• US 10,491,616 B2
• Filed: 02/13/2017
• Issued: 11/26/2019
• Est. Priority Date: 02/13/2017
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Luo, Pengcheng, Briggs, Reeves Hoppe, Sadovsky, Art, Ahmad, Naveed
• Primary Examiner(s): Giddins, Nelson S.
• Application Number:
  US15/431,391
• Publication Number:
  US 20180234442A1
• Time in Patent Office: 1,016 Days
• CPC Class Codes: G06F 21/55 Detecting local intrusion or implementing counter-measures
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: H04L 2463/121 Timestamp
• CPC Class Codes: H04L 63/14 for detecting or protecting against malicious traffic
• CPC Class Codes: H04L 63/1425 Traffic logging, e.g. anomaly detection

Abstract

Detecting compromised devices and user accounts within an online service via multi-signal analysis allows for fewer false positives and thus a more accurate allocation of computing resources and human analyst resources. Individual scopes of analysis, related to devices, accounts, or processes are specified and multiple behaviors over a period of time are analyzed to detect persistent (and slow acting) threats as well as brute force (and fast acting) threats. Analysts are alerted to individually affected scopes suspected of being compromised and may address them accordingly.

Claims

1. A method for detecting compromised scopes in an online service, comprising:

receiving detection results of behaviors occurring on devices within the online service;

scoring each of the detection results based on how anomalous the respective behaviors are within the online service, wherein each of the detection results are scored based on a ratio between a subpopulation count and a population count for the respective behaviors within the online service;

excluding one or more of the scored detection results having one or more of a subpopulation count and a population count for the behaviors below one or more of a subpopulation threshold and a population threshold, respectively;

organizing a remaining one or more of the scored detection results according to scopes;

applying multi-signal detection logic to a given scope from the scopes to produce a confidence score indicating whether the given scope is compromised;

determining whether to present an alert of the given scope being compromised based on comparing the confidence score to an alert threshold; and

in response to determining that the alert is to be presented of the given scope being compromised, generating and transmitting the alert.

2. The method of claim 1, wherein the given scope is associated with a given device or a given user account within the online service.

3. The method of claim 1, wherein scoring each of the detection results further comprises:

computing an identifier for a given detection result;

retrieving an aggregate count for the given detection result;

incrementing aggregate population counts for the given detection result;

calculating an anomaly score for the given detection result based on the aggregate population counts; and

associating the anomaly score with the given detection result to produce a scored detection result.

4. The method of claim 3, wherein the aggregate population counts include:

a raw population count, incremented for each observed detection result; and

at least one subpopulation count, incremented for each observed detection result that includes a characteristic defining a subpopulation of detection results.

5. The method of claim 3, wherein the identifier includes at least one field from the given detection result, including:

a role for a given device associated with the given detection result;

a user associated with the given detection result;

a behavior class associated with the given detection result; and

a timestamp.

6. The method of claim 1, wherein organizing the remaining one or more of the scored detection results according to the scopes further comprises:

receiving a scope definition identifying a characteristic by which to divide the remaining one or more of the scored detection results;

receiving a time window from which the remaining one or more of the scored detection results are to be analyzed; and

dividing the remaining one or more of the scored detection results that were observed within the time window into the scopes according to values of the identified characteristic.

7. The method of claim 6, wherein the characteristic specifies values for:

device identifiers;

user accounts; and

processes identifiers.

8. The method of claim 1, wherein detection results from a smaller window of time, comprising several minutes of detection results, and a larger window of time, comprising several days of detection results, are included in the scopes.

9. The method of claim 1, wherein applying the multi-signal detection logic to the given scope to produce the confidence score further comprises:

selecting a predictive model;

extracting characteristics from a group of the remaining one or more of the scored detection results that were organized according to the given scope;

scoring the characteristics for conversion into numerically valued features; and

providing the features to the predictive model to generate the confidence score.

10. The method of claim 9, wherein the predictive model is generated and selected based on a continuous machine learning process.

11. A system including a processor and a memory storage device storing instructions that when executed by the processor provide for detecting compromised scopes in an online service, comprising:

a signature anomaly scorer, configured to;

receive a given detection result of an event observed occurring in the online service;

determine an extent to which a behavior associated with the given detection result has been previously observed as occurring in the online service by determining a subpopulation count and a population count for the behavior in the online service;

produce an anomaly score to be associated with the given detection result based on a ratio between the subpopulation count and the population count for the behavior; and

determine the subpopulation count and the population count for the behavior is above a subpopulation threshold and a population threshold, respectively; and

a multi-signal detector in communication with the signature anomaly scorer, configured to;

receive one or more detection results sharing a scope in the online service with the given detection result;

receive, from the signature anomaly scorer, the anomaly score associated with the given detection result;

extract features from the given detection result and the one or more detection results sharing the scope, the features including the anomaly score; and

generate a confidence score based on the extracted features for whether the scope is compromised.

12. The system of claim 11, wherein the one or more detection results sharing the scope in the online service with the given detection result are observed within a window of time.

13. The system of claim 12, wherein the extent to which the behavior associated with the given detection result has been previously observed as occurring in the online service is based on occurrences during the window of time.

14. The system of claim 11, wherein the extent to which the behavior associated with the given detection result has been previously observed as occurring in the online service is based on:

a raw population count of events occurring in the online service; and

a raw subpopulation count of a number of occurrences of the behavior in the online service.

15. The system of claim 11, wherein the extent to which the behavior associated with the given detection result has been previously observed as occurring in the online service is based on:

a scoped population count of a number of entities comprising a scope of the online service; and

a scoped subpopulation count of a number of entities in the scope associated with the behavior.

16. The system of claim 15, wherein the entities comprising the scope is defined by an analyst as one of devices within the online service or user accounts of the online service.

17. A computer readable storage device including processor executable instructions for detecting compromised scopes in an online service, comprising:

receiving detection results of behaviors occurring on devices within the online service;

scoring each of the detection results based on how anomalous the respective behaviors are within the online service, wherein each of the detection results are scored based on a ratio between a subpopulation count and a population count for the respective behaviors within the online service;

excluding one or more of the scored detection results having one or more of a subpopulation count and a population count for the respective behaviors below one or more of a subpopulation threshold and a population threshold, respectively;

organizing a remaining one or more of the scored detection results according to scopes, wherein a given scope from the scopes is associated with a given device or a given user account within the online service;

applying multi-signal detection logic to the given scope to produce a confidence score indicating whether the given scope is compromised;

determining whether to alert an analyst of the given scope being compromised based on comparing the confidence score to an alert threshold; and

in response to determining that the analyst is to be alerted of the given scope being compromised, generating and transmitting an alert.

18. The computer readable storage device of claim 17, wherein scoring each of the detection results further comprises:

computing an identifier for a given detection result;

retrieving an aggregate count for the given detection result;

incrementing aggregate population counts for the given detection result;

calculating an anomaly score for the given detection result based on the aggregate population counts; and

associating the anomaly score with the given detection result to produce a scored detection result.

19. The computer readable storage device of claim 17, wherein organizing the remaining one or more of the scored detection results according to the scopes further comprises:

receiving a scope definition identifying a characteristic by which to divide the remaining one or more of the scored detection results;

receiving a time window from which the remaining one or more of the scored detection results are to be analyzed;

dividing the remaining one or more of the scored detection results that were observed within the time window into the scopes according to values of the identified characteristic.

20. The computer readable storage device of claim 17, wherein applying the multi-signal detection logic to the given scope to produce the confidence score further comprises:

selecting a predictive model;

extracting characteristics from a group of the remaining one or more of the scored detection results that were organized according to the given scope;

scoring the characteristics for conversion into numerically valued features; and

providing the features to the predictive model to generate the confidence score.