# US 2021/0334455 A1 · Utility-preserving text de-identification with privacy guarantees

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0334455 A1
- Document ID: US20210334455A1
- Status: Published application
- Kind: A1
- Application number: 16/860,857
- Filed: 2020-04-28
- Published: 2021-10-28
- Earliest claimed: 2020-04-28
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Aris GKOULALAS-DIVANIS; Paul R. Bastide; Xu Wang; Rohit Ranchal
- Web page: https://patentgrep.com/patent/US20210334455A1

## CPC

- G06F 40/151: Transformation
- G06F 40/205: Parsing
- G06F 21/6254: {by anonymising data, e.g. decorrelating personal data from the owner's identification}
- 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)
- G06F 40/169: Annotation, e.g. comment data or footnotes
- G06F 40/216: using statistical methods
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F 40/30: Semantic analysis
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

One embodiment of the invention provides a method for utility-preserving text de-identification. The method comprises generating corresponding processed text for each text document by applying at least one natural language processor (NLP) annotator to the text document to recognize and tag privacy-sensitive personal information corresponding to an individual, and replacing some words in the text document with some replacement values. The method further comprises determining infrequent terms occurring across all processed texts, filtering out the infrequent terms from the processed texts, and selectively reinstating to the processed texts at least one of the infrequent terms that is innocuous. The method further comprises generating a corresponding de-identified text document for each processed text by anonymizing privacy-sensitive personal information corresponding to an individual in the processed text to an extent that preserves data utility of the processed text and conceals the individual's personal identity.

## Claims

### Claim 1

A method for utility-preserving text de-identification, comprising:
for each text document of a set of text documents, generating corresponding processed text by:
applying at least one natural language processor (NLP) annotator to the text document to recognize and tag privacy-sensitive personal information corresponding to an individual; and
replacing at least one word in the text document with at least one replacement value, wherein the at least one word includes a portion of the privacy-sensitive personal information; and
for each processed text corresponding to each text document of the set of text documents:
determining a union of terms appearing in the processed text, wherein the union of terms excludes privacy-sensitive personal information recognized and tagged in the text document;
determining which terms of the union of terms are infrequent;
filtering out each infrequent term from the processed text;
selectively reinstating to the processed text at least one infrequent term that is innocuous; and
generating a corresponding de-identified text document by anonymizing privacy-sensitive personal information in the processed text to an extent that preserves data utility of the processed text and conceals a personal identity of an individual that the privacy-sensitive personal information corresponds to.

### Claim 2 (depends on 1)

The method of claim 1, wherein each privacy-sensitive personal information corresponding to each individual comprises at least one of: a direct identifier, or a quasi-identifier.

### Claim 3 (depends on 2)

The method of claim 2, wherein generating corresponding processed text further comprises:
tagging each identifier recognized in the text document with a set of tags indicative of a type of the identifier;
masking an original value of each direct identifier recognized in the text document with a masked value; and
tagging each quasi-identifier recognized in the text document with another set of tags indicative of a type of the quasi-identifier.

### Claim 4 (depends on 3)

The method of claim 3, further comprising:
for each processed text corresponding to each text document of the set of text documents:
determining, for each term of the union of terms appearing in the processed text, a corresponding term frequency indicative of a number of text documents from the set of text documents that contain the term; and
determining which terms of the union of terms are infrequent based on each term frequency corresponding to each term of the union of terms and a frequency threshold derived from a blacklist, wherein each infrequent term has a corresponding term frequency that is less than the frequency threshold.

### Claim 5

(canceled)

### Claim 6 (depends on 4)

The method of claim 4, further comprising:
for each processed text corresponding to each text document of the set of text documents:
for each infrequent term filtered out of the processed text, determining whether the infrequent term is innocuous based on a whitelist, and reinstating the infrequent term to the processed text in response to determining the infrequent term is innocuous.

### Claim 7 (depends on 6)

The method of claim 6, further comprising:
for each processed text, generating one or more corresponding records of structured data, wherein each corresponding record maintains one or more original values for one or more quasi-identifiers tagged in the processed text, and the one or more quasi-identifiers are included in a list of known quasi-identifiers.

### Claim 8 (depends on 7)

The method of claim 7, wherein anonymizing privacy-sensitive personal information included in the processed text comprises:
anonymizing one or more original values maintained in one or more corresponding records, and replacing the one or more original values in the processed text with one or more resulting anonymized values.

### Claim 9 (depends on 8)

The method of claim 8, wherein anonymizing privacy-sensitive personal information included in the processed text further comprises:
for each identifier tagged in the processed text, removing each tag that the identifier is tagged with from the processed text.

### Claim 10

A system for utility-preserving text de-identification, comprising:
at least one processor; and
a non-transitory processor-readable memory device storing instructions that when executed by the at least one processor causes the at least one processor to perform operations including:
for each text document of a set of text documents, generating corresponding processed text by:
applying at least one natural language processor (NLP) annotator to the text document to recognize and tag privacy-sensitive personal information corresponding to an individual; and
replacing at least one word in the text document with at least one replacement value, wherein the at least one word includes a portion of the privacy-sensitive personal information; and
for each processed text corresponding to each text document of the set of text documents:
determining a union of terms appearing in the processed text, wherein the union of terms excludes privacy-sensitive personal information recognized and tagged in the text document;
determining which terms of the union of terms are infrequent;
filtering out each infrequent term from the processed text;
selectively reinstating to the processed text at least one infrequent term that is innocuous; and
generating a corresponding de-identified text document by anonymizing privacy-sensitive personal information in the processed text to an extent that preserves data utility of the processed text and conceals a personal identity of an individual that the privacy-sensitive personal information corresponds to.

### Claim 11 (depends on 10)

The system of claim 10, wherein each privacy-sensitive personal information corresponding to each individual comprises at least one of: a direct identifier, or a quasi-identifier.

### Claim 12 (depends on 11)

The system of claim 11, wherein generating corresponding processed text further comprises:
tagging each identifier recognized in the text document with a set of tags indicative of a type of the identifier;
masking an original value of each direct identifier recognized in the text document with a masked value; and
tagging each quasi-identifier recognized in the text document with another set of tags indicative of a type of the quasi-identifier.

### Claim 13 (depends on 12)

The system of claim 12, wherein the operations further comprise:
for each processed text corresponding to each text document of the set of text documents:
determining, for each term of the union of terms appearing in the processed text, a corresponding term frequency indicative of a number of text documents from the set of text documents that contain the term; and
determining which terms of the union of terms are infrequent based on each term frequency corresponding to each term of the union of terms and a frequency threshold derived from a blacklist, wherein each infrequent term has a corresponding term frequency that is less than the frequency threshold.

### Claim 14

(canceled)

### Claim 15 (depends on 13)

The system of claim 13, wherein the operations further comprise:
for each processed text corresponding to each text document of the set of text documents:
for each infrequent term filtered out of the processed text, determining whether the infrequent term is innocuous based on a whitelist, and reinstating the infrequent term to the processed text in response to determining the infrequent term is innocuous.

### Claim 16 (depends on 15)

The system of claim 15, wherein the operations further comprise:
for each processed text, generating one or more corresponding records of structured data, wherein each corresponding record maintains one or more original values for one or more quasi-identifiers tagged in the processed text, and the one or more quasi-identifiers are included in a list of known quasi-identifiers.

### Claim 17 (depends on 16)

The system of claim 16, wherein anonymizing privacy-sensitive personal information included in the processed text comprises:
anonymizing one or more original values maintained in one or more corresponding records, and replacing the one or more original values in the processed text with one or more resulting anonymized values.

### Claim 18 (depends on 17)

The system of claim 17, wherein anonymizing privacy-sensitive personal information included in the processed text further comprises:
for each identifier tagged in the processed text, removing each tag that the identifier is tagged with from the processed text.

### Claim 19

A computer program product for utility-preserving text de-identification, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
for each text document of a set of text documents, generating corresponding processed text by:
applying at least one natural language processor (NLP) annotator to the text document to recognize and tag privacy-sensitive personal information corresponding to an individual; and
replacing at least one word in the text document with at least one replacement value, wherein the at least one word includes a portion of the privacy-sensitive personal information; and
for each processed text corresponding to each text document of the set of text documents:
determining a union of terms appearing in the processed text, wherein the union of terms excludes privacy-sensitive personal information recognized and tagged in the text document;
determining which terms of the union of terms are infrequent;
filtering out each infrequent term from the processed text;
selectively reinstating to the processed text at least one infrequent term that is innocuous; and
generating a corresponding de-identified text document by anonymizing privacy-sensitive personal information in the processed text to an extent that preserves data utility of the processed text and conceals a personal identity of an individual that the privacy-sensitive personal information corresponds to.

### Claim 20 (depends on 19)

The computer program product of claim 19, wherein each privacy-sensitive personal information corresponding to each individual comprises at least one of: a direct identifier, or a quasi-identifier.
