Differential privacy with cloud data

• US 11,496,286 B2
• Filed: 11/16/2017
• Issued: 11/08/2022
• Est. Priority Date: 01/08/2017
• Status: Active Grant

• Current Assignee: Apple Inc.
• Original Assignee: Apple Inc.
• Inventors: Duffy, William T., Vyrros, Andrew H., Minadakis, Yannis, Byde, Andrew R., Pagallo, Giulia
• Primary Examiner(s): Najjar, Saleh
• Assistant Examiner(s): Nguyen, Nhan Huu
• Application Number:
  US15/815,611
• Publication Number:
  US 20180198602A1
• Time in Patent Office: 1,818 Days
• CPC Class Codes: G06F 21/6245 Protecting personal data, e.g. for financial or medical purposes
• CPC Class Codes: G06F 2221/2107 File encryption
• CPC Class Codes: G06F 40/117 Tagging; Marking up details of markup languages G06F40/143; Designating a block; Setting of attributes style sheets, e.g. eXtensible Stylesheet Language Transformation [XSLT], G06F40/154
• CPC Class Codes: G06F 40/20 Natural language analysis semantic analysis of natural language G06F40/30
• CPC Class Codes: H04L 9/008 involving homomorphic encryption
• CPC Class Codes: H04L 9/0816 Key establishment, i.e. cryptographic processes or cryptographic protocols whereby a shared secret becomes available to two or more parties, for subsequent use

Abstract

Embodiments described herein enable data associated with a large plurality of users to be analyzed without compromising the privacy of the user data. In one embodiment, a user can opt-in to allow analysis of clear text of the user'"'"'s emails. An analysis process can then be performed in which an analysis service receives clear text of an email of a client device; processes the clear text of the email into one or more tokens having one or more tags; enriches one or more tokens in the processed email using data associated with a user of the client device and the one or more tags; and processes the clear text and one or more enriched tokens to generate a data set of one or more feature vectors.

Claims

1. A computer-implemented method comprising:

receiving, by a service, clear text of an email of a client device;

processing, by the service, the clear text of the email into a processed email, wherein processing the clear text of the email includes parsing the clear text into one or more tokenized sentence segments including one or more tags, and wherein the one or more tags privatize personally identifiable information within the clear text of the email;

enriching the one or more tokenized sentence segments in the processed email using data from an application specific to a user of the client device and the one or more tags, wherein enriching the tokenized sentence segments comprises;

accessing, from the application on the client device, the data specific to the user of the client device;

comparing the clear text associated with the one or more tags with the data associated with the user of the client device; and

in response to determining a match between the clear text and the data associated with the user of the client device;

replacing, by the service, at least one of the one or more tags with an enriched tag, wherein the enriched tag is created by modifying the at least one or more tags to include the data from the application specific to the user of the client device.

2. The method of claim 1, further comprising:

in response to receiving the email, encrypting the clear text of the email into an encrypted email;

storing the encrypted email; and

decrypting the encrypted email before processing the clear text of the email.

3. The method of claim 1, wherein the user of the client device has opted-in to usage, by the service, of emails of the user, and the data associated with the user of the client device comprises one or more of a calendar and a contacts list.

4. The method of claim 1, further comprising:

processing the clear text, the one or more tags, and the enriched tag to generate a data set of one or more feature vectors;

encrypting the data set of one or more feature vectors;

generating a privatized data set from the encrypted data set, wherein the privatized data set includes a differentially private sketch for each feature vector;

accumulating the differentially private sketches to obtain a frequency of each differentially private feature vector over a batch of processed feature vector sets; and

discarding one or more differentially private feature vectors having an accumulated frequency less than a threshold.

5. The method of claim 4, further comprising:

selecting one or more encrypted feature vectors from the privatized data set;

decrypting the selected, encrypted one or more feature vectors; and

transmitting the decrypted one or more feature vectors as the data set of feature vectors to the client device within an asset catalog update for the client device.

6. The method of claim 5, wherein the asset catalog includes a feature vector to enable the client device to automatically generate a contact entry in response to detecting that the user has typed text matching the feature vector.

7. The method of claim 4, wherein the data set of feature vectors is used by a natural language processing feature of an application.

8. The method of claim 4, wherein the data set of feature vectors is used by a predictive text feature of an application.

9. The method according to claim 1, wherein the application on the client device that is specific to the user comprises a user calendar or a user contacts database.

10. A non-transitory computer readable medium programmed with instructions that, when executed by a processing system, cause the processing system to perform operations comprising:

receiving, by a service, clear text of an email of a client device;

processing, by the service, the clear text of the email into a processed email, wherein processing the clear text of the email includes parsing the clear text into one or more tokenized sentence segments including one or more tags, wherein the one or more tags privatize personally identifiable information within the clear text of the email;

enriching the one or more tokenized sentence segments in the processed email using data from an application specific to a user of the client device and the one or more tags, wherein enriching the tokenized sentence segments comprises;

accessing, from the application on the client device, the data specific to the user of the client device;

comparing the clear text associated with the one or more tags with the data associated with the user of the client device; and

in response to determining a match between the clear text and the data associated with the user of the client device, replacing, by the service, at least one of the one or more tags with an enriched tag, wherein the enriched tag is created by modifying the at least one or more tags to include the data from the application specific to the user of the client device.

11. The non-transitory computer readable medium of claim 10, the operations further comprising:

encrypting the clear text of the email, in response to receiving the email, the clear text of the email encrypted into an encrypted email;

storing the encrypted email; and

decrypting the encrypted email before processing the clear text of the email.

12. The non-transitory computer readable medium of claim 10, the user of the client device having opted-in to usage, by the service, of data associated with the user of the client device, the data associated with the user of the client device including e-mail and one or more of:

a calendar and a contacts list.

13. The non-transitory computer readable medium of claim 10, the operations further comprising:

processing the clear text, the one or more tags and the enriched tag to generate a data set of one or more feature vectors;

encrypting the data set of one or more feature vectors;

generating a privatized data set from the encrypted data set, wherein the privatized data set includes a differentially private sketch for each feature vector;

accumulating the differentially private sketches to obtain a frequency of each differentially private feature vector over a batch of processed feature vector sets; and

discarding one or more differentially private feature vectors having an accumulated frequency less than a threshold

selecting one or more encrypted feature vectors from the privatized data set;

decrypting the selected, encrypted one or more feature vectors; and

transmitting the decrypted one or more feature vectors as the data set of feature vectors to the client device within an asset catalog update for the client device.

14. The non-transitory computer readable medium of claim 13, wherein the asset catalog includes a feature vector to enable the client device to automatically generate a contact entry in response to detecting that the user has typed text matching the feature vector.

15. The non-transitory computer readable medium of claim 13, wherein the data set of feature vectors is used by a natural language processing feature of an application or a predictive text feature of an application.

16. A system comprising:

a non-transitory machine-readable medium storing instructions;

a processing system to execute instructions stored on the non-transitory machine-readable medium, the instructions to cause the processing system to;

receive clear text of an email of a client device;

process the clear text of the email into a processed email, the processed email including one or more tokens having one or more tags, wherein to process the clear text of the email includes to parse the clear text into one or more tokenized sentence segments including one or more tags, wherein the one or more tags privatize personally identifiable information within the clear text of the email;

enrich the one or more tokenized sentence segments in the processed email using data from an application specific to a user of the client device and the one or more tags, wherein to enrich the tokenized sentence segments comprises;

accessing, from the application on the client device, the data specific to the user of the client device;

comparing the clear text associated with the one or more tags with the data associated with the user of the client device; and

in response to determining a match between the clear text and the data associated with the user of the client device, replacing at least one of the one or more tags with an enriched tag, wherein the enriched tag is created by modifying the at least one or more tags to include the data from the application specific to the user of the client device.

17. The system of claim 16, the processing system further to:

encrypt the clear text of the email, in response to receiving the email;

store the encrypted email; and

decrypt the encrypted email before processing the clear text of the email.

18. The system of claim 16, the user of the client device having opted-in to usage of data associate with the user, the data associated with the user of the client device including email and one or more of a calendar and a contacts list.

19. The system of claim 16, the processing system further to:

process the clear text, the one or more tags and enriched tag to generate a data set of one or more feature vectors;

encrypt the data set of one or more feature vectors;

generate a privatized data set from the one or more feature vectors, wherein

the privatized data set includes a differentially private sketch for each feature vector;

accumulate the differentially private sketches to obtain a frequency of each differentially private feature vector over a batch of processed feature vector sets;

discard one or more differentially private feature vectors having an accumulated frequency less than a threshold;

select one or more encrypted feature vectors from the privatized data set;

decrypt the encrypted one or more feature vectors into decrypted one or more feature vectors; and

transmit the decrypted one or more feature vectors as the data set of feature vectors to the client device within an asset catalog update for the client device.

20. The system as in claim 19, wherein the asset catalog includes a feature vector to enable the client device to automatically generate a contact entry in response to detecting that the user has typed text matching the feature vector.

21. The system of claim 19, the data set of feature vectors to be used by a natural language processing feature of an application or a predictive text feature of an application.