Identifying cohorts with anomalous confidential data submissions using matrix factorization and completion techniques

• US 10,037,437 B1
• Filed: 06/09/2017
• Issued: 07/31/2018
• Est. Priority Date: 06/09/2017
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Kenthapadi, Krishnaram, Ambler, Stuart MacDonald
• Primary Examiner(s): Nipa, Wasika
• Application Number:
  US15/618,583
• Time in Patent Office: 417 Days
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06F 21/6227 where protection concerns the structure of data, e.g. records, types, queries
• CPC Class Codes: G06F 21/6245 Protecting personal data, e.g. for financial or medical purposes
• CPC Class Codes: G06F 21/6254 by anonymising data, e.g. decorrelating personal data from the owner's identification
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 50/01 Social networking
• CPC Class Codes: H04L 9/0897 involving additional devices, e.g. trusted platform module [TPM], smartcard or USB
• CPC Class Codes: H04L 9/14 using a plurality of keys or algorithms

Abstract

In an example, for each value of a plurality of values of a first attribute of members of a social networking service who have submitted confidential data, an allowed range for normalized confidential data values submitted by members having the value for the first attribute, across all values of a second attribute, is calculated, and then shifted based on an inferred median confidential data value relative to a median of confidential data values. Then, anomalous confidential data values can be detected using this information.

Claims

1. A system comprising:

one or more hardware processors;

a non-transitory computer-readable medium having instructions stored thereon, which, when executed by the one or more hardware processors, cause the system to;

for each value of a plurality of values of a first attribute of members of a social networking service who have submitted confidential data;

calculate, using the one or more hardware processors, an allowed range for confidential data values submitted by members having the value for the first attribute, across all values of a second attribute, the confidential data values having been entered on screens of graphical user interfaces and encrypted on an external data source;

calculate, using the one or more hardware processors, a median for the confidential data values by selecting a midpoint in a distribution of the confidential data values;

identify a set of candidate data transformation functions, each candidate data transformation function applying a different function on the confidential data values;

shift, using the one or more hardware processors, the allowed range based on an inferred median confidential data value relative to the median of the confidential data values, the inferred median confidential data value being inferred by optimizing parameters of an objective function that minimize an error function for the objective function to select one candidate data transformation functions from the set of candidate data transformation functions, applying the selected one candidate data transformation function to the confidential data values to transform the confidential data values, and calculating a median for the transformed confidential data values;

for each value of the plurality of values of the first attribute of members of the social networking service who have submitted confidential data and each value of a plurality of values of the second attribute of members of the social networking service who have submitted confidential data;

determine, using the one or more hardware processors, whether the submitted confidential data from members having the value of the first attribute and the value of the second attribute is outside the shifted allowed range for the value of the first attribute; and

in response to a determination that the submitted confidential data from members having the value of the first attribute and the value of the second attribute is outside the shifted allowed range for the value of the first attribute, mark, using the one or more hardware processors, a combination of the value of the first attribute and the value of the second attribute as anomalous.

2. The system of claim 1, wherein the calculating an allowed range includes:

computing an empirical distribution of the confidential data values for the value of the first attribute;

computing an upper and lower end of the allowed range from the empirical distribution; and

calculating a median for the allowed range from the empirical distribution.

3. The system of claim 1, wherein the shifting includes inferring a median confidential data value using matrix factorization.

4. The system of claim 3, wherein the inferring comprises:

constructing a matrix of the confidential data values having the first attribute as a first axis and the second attribute as a second axis, with each cell in the matrix corresponding to corresponding different combinations of values of the first attribute and the second attribute;

computing a set of candidate low rank approximations of the matrix using an objective function evaluated using a set of candidate data transformation functions, the objective function having one or more parameters and an error function;

optimizing the one or more parameters that minimizes the error function of the objective function to select one of the candidate low rank approximations of the matrix; and

inferring one or more cells that are missing data, of the selected one of the candidate low rank approximations of the matrix.

5. The system of claim 4, wherein the constructing the matrix includes using a machine learning model to generate median confidential data values for one or more cells along with corresponding confidence scores.

6. The system of claim 4, wherein the computing the set of candidate low rank approximations includes, for each candidate data transformation:

applying the candidate data transformation to the matrix;

obtaining a training matrix by hiding a preset fraction of entries of the transformed matrix; and

for each of one or more candidate parameter values for one of the one or more parameters;

computing the objective function using the candidate parameter value; and

calculating the error function using the candidate parameter value.

7. The system of claim 4, wherein the inferring the one or more cells that are missing data includes:

for each cell that is missing data, of the selected one of the candidate low rank approximations of the matrix;

determining if the cell that is missing data is present in the matrix of the confidential data values; and

in response to a determination that the cell that is missing data is present in the matrix of confidential data values, applying an inverse of the selected candidate low rank approximations to the cell that is missing data.

8. A method comprising:

for each value of a plurality of values of a first attribute of members of a social networking service who have submitted confidential data;

calculating, using one or more hardware processors, an allowed range for confidential data values submitted by members having the value for the first attribute, across all values of a second attribute, the confidential data values having been entered on screens of graphical user interfaces and encrypted on an external data source;

calculating, using the one or more hardware processors, a median for the confidential data values by selecting a midpoint in a distribution of the confidential data values;

identifying a set of candidate data transformation functions, each candidate data transformation function applying a different function on the confidential data values;

shifting, using the one or more hardware processors, the allowed range based on an inferred median confidential data value relative to the median of the confidential data values, the inferred median confidential data value being inferred by optimizing parameters of an objective function that minimize an error function for the objective function to select one candidate data transformation functions from the set of candidate data transformation functions, applying the selected one candidate data transformation function to the confidential data values to transform the confidential data values, and calculating a median for the transformed confidential data values;

for each value of the plurality of values of the first attribute of members of the social networking service who have submitted confidential data and each value of a plurality of values of the second attribute of members of the social networking service who have submitted confidential data;

determining, using the one or more hardware processors, whether the submitted confidential data from members having the value of the first attribute and the value of the second attribute is outside the shifted allowed range for the value of the first attribute; and

in response to a determination that the submitted confidential data from members having the value of the first attribute and the value of the second attribute is outside the shifted allowed range for the value of the first attribute, marking, using the one or more hardware processors, a combination of the value of the first attribute and the value of the second attribute as anomalous.

9. The method of claim 8, wherein the calculating an allowed range includes:

computing an empirical distribution of the confidential data values for the value of the first attribute;

computing an upper and lower end of the allowed range from the empirical distribution; and

calculating a median for the allowed range from the empirical distribution.

10. The method of claim 8, wherein the shifting includes inferring a median confidential data value using matrix factorization.

11. The method of claim 10, wherein the inferring comprises:

constructing a matrix of the confidential data values having the first attribute type as a first axis and the second attribute type as a second axis, with each cell in the matrix corresponding to corresponding different combinations of values of the first attribute and the second attribute;

computing a set of candidate low rank approximations of the matrix using an objective function evaluated using a set of candidate data transformation functions, the objective function having one or more parameters and an error function;

optimizing the one or more parameters that minimizes the error function of the objective function to select one of the candidate low rank approximations of the matrix; and

inferring one or more cells that are missing data, of the selected one of the candidate low rank approximations of the matrix.

12. The method of claim 10, wherein the constructing the matrix includes using a machine learning model to generate median confidential data values for one or more cells along with corresponding confidence scores.

13. The method of claim 10, wherein the computing a set of candidate low rank approximations includes, for each candidate data transformation:

applying the candidate data transformation to the matrix;

obtaining a training matrix by hiding a preset fraction of entries of the transformed matrix; and

for each of one or more candidate parameter values for one of the one or more parameters;

computing the objective function using the candidate parameter value; and

calculating the error function using the candidate parameter value.

14. The method of claim 11, wherein the inferring the one or more cells that are missing data includes:

for each cell that is missing data, of the selected one of the candidate low rank approximations of the matrix;

determining if the cell that is missing data is present in the matrix of the confidential data values; and

in response to a determination that the cell that is missing data is present in the matrix of confidential data values, applying an inverse of the selected candidate low rank approximations to the cell that is missing data.

15. A non-transitory machine-readable storage medium comprising instructions, which when implemented by one or more machines, cause the one or more machines to perform operations comprising:

for each value of a plurality of values of a first attribute of members of a social networking service who have submitted confidential data;

calculating, using one or more hardware processors, an allowed range for confidential data values submitted by members having the value for the first attribute, across all values of a second attribute, the confidential data values having been entered on screens of graphical user interfaces and encrypted on an external data source;

calculating, using the one or more hardware processors, a median for the confidential data values by selecting a midpoint in a distribution of the confidential data values;

identifying a set of candidate data transformation functions, each candidate data transformation function applying a different function on the confidential data values;

shifting, using the one or more hardware processors, the allowed range based on an inferred median confidential data value relative to the median of the confidential data values, the inferred median confidential data value being inferred by optimizing parameters of an objective function that minimize an error function for the objective function to select one candidate data transformation functions from the set of candidate data transformation functions, applying the selected one candidate data transformation function to the confidential data values to transform the confidential data values, and calculating a median for the transformed confidential data values;

for each value of the plurality of values of the first attribute of members of the social networking service who have submitted confidential data and each value of a plurality of values of the second attribute of members of the social networking service who have submitted confidential data;

determining, using the one or more hardware processors, whether the submitted confidential data from members having the value of the first attribute and the value of the second attribute is outside the shifted allowed range for the value of the first attribute; and

in response to a determination that the submitted confidential data from members having the value of the first attribute and the value of the second attribute is outside the shifted allowed range for the value of the first attribute, marking, using the one or more hardware processors, a combination of the value of the first attribute and the value of the second attribute as anomalous.

16. The non-transitory machine-readable storage medium of claim 15, wherein the calculating an allowed range includes:

computing an empirical distribution of the confidential data values for the value of the first attribute;

computing an upper and lower end of the allowed range from the empirical distribution; and

calculating a median for the allowed range from the empirical distribution.

17. The non-transitory machine-readable storage medium of claim 15, wherein the shifting includes inferring a median confidential data value using matrix factorization.

18. The non-transitory machine-readable storage medium of claim 15, wherein the inferring comprises:

constructing a matrix of the confidential data values having the first attribute type as a first axis and the second attribute type as a second axis, with each cell in the matrix corresponding to corresponding different combinations of values of the first attribute and the second attribute;

computing a set of candidate low rank approximations of the matrix using an objective function evaluated using a set of candidate data transformation functions, the objective function having one or more parameters and an error function;

optimizing the one or more parameters that minimizes the error function of the objective function to select one of the candidate low rank approximations of the matrix; and

inferring one or more cells that are missing data, of the selected one of the candidate low rank approximations of the matrix.

19. The non-transitory machine-readable storage medium of claim 18, wherein the constructing the matrix includes using a machine learning model to generate median confidential data values for one or more cells along with corresponding confidence scores.

20. The non-transitory machine-readable storage medium of claim 18, wherein the computing a set of candidate low rank approximations includes, for each candidate data transformation:

applying the candidate data transformation to the matrix;

obtaining a training matrix by hiding a preset fraction of entries of the transformed matrix; and

for each of one or more candidate parameter values for one of the one or more parameters;

computing the objective function using the candidate parameter value; and

calculating the error function using the candidate parameter value.