# US 2019/0377905 A1 · Mixed Data Fingerprinting with Principal Components Analysis

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

- Number: US 2019/0377905 A1
- Document ID: US20190377905A1
- Status: Published application
- Kind: A1
- Application number: 16/472,295
- Filed: 2017-11-20
- Published: 2019-12-12
- Earliest claimed: 2016-12-22
- Assignees: LiveRamp, Inc.
- Applicants: LiveRamp, Inc.
- Inventors: Martin Rose; Christina Tsz Ling Leung
- Web page: https://patentgrep.com/patent/US20190377905A1

## CPC

- G06F 21/6218: {to a system of files or objects, e.g. local or distributed file system or database}
- G06F 18/21345: {enforcing sparsity or involving a domain transformation}
- G06F 18/2135: based on approximation criteria, e.g. principal component analysis
- G06F 21/00: Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F 21/16: Program or content traceability, e.g. by watermarking
- G06F 21/64: Protecting data integrity, e.g. using checksums, certificates or signatures
- H04L 9/3247: {involving digital signatures}
- G06F 16/17: Details of further file system functions
- G06F 16/41: Indexing; Data structures therefor; Storage structures
- G06F 16/61: Indexing; Data structures therefor; Storage structures
- G06F 16/68: Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F 17/12: Simultaneous equations {, e.g. systems of linear equations}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06F 21/552: {involving long-term monitoring or reporting}
- G06F 21/577: {Assessing vulnerabilities and evaluating computer system security}
- G06F 21/62: Protecting access to data via a platform, e.g. using keys or access control rules
- H04L 2209/608: Watermarking
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

Principal components analysis is applied to data sets to fingerprint the dataset or to compare the dataset to a “wild file” that may have been constructed from data found in the dataset. Principal components analysis allows for the reduction of data used for comparison down to a parsimonious compressed signature of a dataset. Datasets with different patterns among the variables will have different patterns of principal components. The principal components of variables (or a relevant subset thereof) in a wild file may be computed and statistically compared to the principal components of identical variables in a data provider's reference file to provide a score. This constitutes a unique and compressed signature of a file that can be used for identification and comparison with similarly defined patterns from other files.

## Claims

### Claim 1

An apparatus for determining whether a wild file was leaked from a data owner dataset, wherein the data owner dataset and the wild file each comprise a plurality of records each comprising a plurality of fields, the apparatus comprising:
a. a subset selection subroutine, wherein the subset selection subroutine is configured to receive a data owner dataset and a date-adjusted wild file and create a data owner subset and a wild file subset, wherein the data owner subset and the wild file subset contain a reduced number of records compared to the data owner dataset and the wild file, respectively;
b. a PCAmix data analysis subroutine, wherein the PCAmix data analysis subroutine is configured to receive the data owner subset and the wild file subset and produce a set of data owner subset eigenvalues, a set of data owner subset eigenvectors, a set of wild file subset eigenvalues, and a set of wild file subset eigenvectors; and
c. a score generation subroutine, wherein the score generation subroutine is configured to receive the data owner subset eigenvalues, the data owner subset eigenvectors, the wild file subset eigenvalues, and the wild file subset eigenvectors to produce a similarity score indicative of the likelihood that the wild file was derived from the data owner dataset.

### Claim 2 (depends on 1)

The apparatus of claim 1, further comprising a file date determination subroutine configured to analyze the wild file and adjust dates in the wild file to account for a passage of time since the wild file was leaked to produce a date-adjusted wild file.

### Claim 3 (depends on 2)

The apparatus of claim 2, further comprising a record matching routine configured to receive the date-adjusted wild file and the data owner dataset and produce a set of matched records between the date-adjusted wild file and the data owner dataset.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the PCAmix data analysis subroutine is further configured to build a numerical data matrix, a first diagonal matrix constructed from row weights from the numerical data matrix, and a second diagonal matrix constructed from column weights from the numerical data matrix.

### Claim 5 (depends on 4)

The apparatus of claim 4, wherein the PCAmix data analysis subroutine is further configured to perform a generalized singular value decomposition on the numerical data matrix using metrics from the first and second diagonal matrices to produce the data set owner subset eigenvectors and the wild file subset eigenvectors.

### Claim 6 (depends on 5)

The apparatus of claim 5, wherein the PCAmix data analysis subroutine is further configured to eliminate all of the data set owner subset eigenvectors and wild file subset eigenvectors other than those that account for a significant portion of variance.

### Claim 7 (depends on 6)

The apparatus of claim 6, wherein the PCAmix data analysis subroutine is further configured to eliminate all of the data set owner subset eigenvectors and wild file subset eigenvectors other than those that account for at least ten percent of variance.

### Claim 8

An apparatus for creating a fingerprint for a data owner dataset, wherein the data owner dataset comprises a plurality of records each comprising a plurality of fields, the apparatus comprising:
a. a subset selection subroutine, wherein the subset selection subroutine is configured to receive a data owner dataset and create from the data owner dataset a data owner subset, wherein the data owner subset contains a reduced number of records compared to the data owner dataset; and
b. a PCAmix data analysis subroutine, wherein the PCAmix data analysis subroutine is configured to receive the data owner subset and produce a set of data owner subset eigenvalues and a set of data owner subset eigenvectors, a set of wild file eigenvalues.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein the PCAmix data analysis subroutine is further configured to build a numerical data matrix, a first diagonal matrix constructed from a set of row weights from the numerical data matrix, and a second diagonal matrix constructed from a set of column weights from the numerical data matrix.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the PCAmix data analysis subroutine is further configured to perform a generalized singular value decomposition on the numerical data matrix using a set of metrics from the first and second diagonal matrices to produce the data set owner subset eigenvectors.

### Claim 11 (depends on 10)

The apparatus of claim 10, wherein the PCAmix data analysis subroutine is further configured to eliminate all of the data set owner subset eigenvectors other than those that account for a significant portion of variance.

### Claim 12 (depends on 11)

The apparatus of claim 11, wherein the PCAmix data analysis subroutine is further configured to eliminate all of the data set owner subset eigenvectors other than those that account for at least ten percent of variance.

### Claim 13

A method for fingerprinting a data owner dataset, wherein the dataset comprises a plurality of records each comprising a plurality of fields, the method comprising the steps of:
a. selecting a subset of the records from the dataset to produce a data owner subset;
b. applying principal components analysis to the data owner subset to produce a matrix of data owner subset eigenvalues and a matrix of data owner subset eigenvectors; and
c. analyzing the eigenvectors to produce a set of scores that define observational values on the principal components of the data owner subset.

### Claim 14 (depends on 13)

The method of claim 13, wherein the step of applying principal components analysis to the data owner subset further comprises the step of building a numerical data matrix, a first diagonal matrix constructed from a set of row weights from the numerical data matrix, and a second diagonal matrix constructed from a set of column weights from the numerical data matrix.

### Claim 15 (depends on 14)

The method of claim 14, wherein the step of applying principal components analysis to the data owner subset further comprises the step of performing a generalized singular value decomposition on the numerical data matrix using a set of metrics from the first and second diagonal matrices to produce the data set owner subset eigenvectors.

### Claim 16 (depends on 15)

The method of claim 15, wherein the step of applying principal components analysis to the data owner subset further comprises the step of removing all of the data set owner subset eigenvectors other than those that account for a significant portion of variance.

### Claim 17 (depends on 16)

The method of claim 16, wherein the step of applying principal components analysis to the data owner subset further comprises the step of removing all of the data set owner subset eigenvectors other than those that account for at least ten percent of variance.

### Claim 18

A method for determining if a wild file is derived from a data owner dataset, wherein each of the data owner dataset and the wild file comprise a plurality of records each comprising a plurality of fields, the method comprising the steps of:
a. extracting a subset of the records from the data owner dataset to produce a data owner subset;
b. extracting a subset of the records from the wild file corresponding to the records in the data owner subset to produce a wild file subset;
c. applying principal components analysis to the data owner subset to produce a set of data owner subset eigenvalues and a matrix of data owner subset eigenvectors;
d. applying principal components analysis to the wild file subset to produce a set of wild file subset eigenvalues and a matrix of wild file subset eigenvectors;
e. analyzing each data owner subset eigenvector relative to a corresponding wild file subset eigenvector to produce a similarity score.

### Claim 19 (depends on 18)

The method of claim 18, further comprising the step of determining a file date for the wild file prior to the step of selecting a subset of the records from the wild file.

### Claim 20 (depends on 18)

The method of claim 18, wherein the step of applying principal components analysis to the data owner subset further comprises the step of building a numerical data matrix, a first diagonal matrix constructed from row weights from the numerical data matrix, and a second diagonal matrix constructed from column weights from the numerical data matrix.

### Claim 21 (depends on 20)

The method of claim 20, wherein the step of applying principal components analysis to the data owner subset further comprises the step of performing a generalized singular value decomposition on the numerical data matrix using metrics from the first and second diagonal matrices to produce the data set owner subset eigenvectors and the wild file subset eigenvectors.

### Claim 22 (depends on 21)

The method of claim 21, wherein the step of applying principal components analysis to the data owner subset further comprises the step of eliminating all of the data set owner subset eigenvectors and wild file subset eigenvectors other than those that account for a significant portion of variance.

### Claim 23 (depends on 22)

The method of claim 22, wherein the step of applying principal components analysis to the data owner subset further comprises the step of eliminating all of the data set owner subset eigenvectors and wild file subset eigenvectors other than those that account for at least ten percent of variance.
