# US 2021/0263793 A1 · Entity resolution framework for data matching

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

- Number: US 2021/0263793 A1
- Document ID: US20210263793A1
- Status: Published application
- Kind: A1
- Application number: 17/208,569
- Filed: 2021-03-22
- Published: 2021-08-26
- Earliest claimed: 2018-12-11
- Assignees: Rovi Guides, Inc.
- Applicants: Rovi Guides, Inc.
- Inventors: Michael James Brehm
- Web page: https://patentgrep.com/patent/US20210263793A1

## CPC

- G06F 11/0793: {Remedial or corrective actions (recovery from an exception in an instruction pipeline G06F9/3861; by retry G06F11/1402; for recovering from a failure of a protocol instance or entity H04L69/40)}
- G06F 16/215: Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
- G06F 11/0727: {in a storage system, e.g. in a DASD or network based storage system (drivers for digital recording or reproducing units G06F3/06; circuits for error detection or correction within digital recording or reproducing units G11B20/18; for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS], H04L67/1097)}
- G06F 11/1474: in transactions (updating of structured data in databases G06F16/23)
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/0499: Feedforward networks
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/09: Supervised learning
- G06F 40/216: using statistical methods
- G06F 40/226: Validation
- G06F 2201/80: Database-specific techniques
- G06N 3/045: Combinations of networks
- G06N 20/00: Machine learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Systems and methods are described for matching a corrupted database record with a record of a validated database. The system receives a corrupted record from a first database. The corrupted record is vectorized to create an input data vector. A denoised data vector is generated by applying a denoising autoencoder to the input data vector, where the denoising autoencoder is specific to the first database. The system compares the denoised data vector with each of a plurality of validated data vectors generated based on records of the validated database to determine that a first denoised data vector matches a matching vector. In response, the system trains the denoising autoencoder using a data pair that includes the input data vector and the matching vector. The system also outputs the validated record that was used to generate the first matching vector.

## Claims

### Claim at position 1

1-30. (canceled)

### Claim 31

A method comprising:
receiving a corrupted record from a first database;
generating a first input data vector based on the corrupted record using a denoising autoencoder, wherein the denoising autoencoder was trained using a plurality of records from the first database;
generating a second data vector based on a validated record from a validated database; and
in response to determining a match based on comparing the first input data vector with the second data vector:
transmitting the validated record to the first database.

### Claim 32 (depends on 31)

The method of claim 31, wherein, in response to determining that a match does not exist, outputting an indication of a match failure.

### Claim 33 (depends on 31)

The method of claim 31, in response to determining the match:
determining that the corrupted record is incomplete;
causing the first database to complete the corrupted record based on the validated record.

### Claim 34 (depends on 31)

The method of claim 31, further comprising:
using the additional training example data pair as an input to a neural network;
generating an output based on the additional training example data pair as the input; and
evaluating the neural network's performance based on the output.

### Claim 35 (depends on 34)

The method of claim 34, wherein training the denoising autoencoder using the plurality of records from the first database comprises adjusting neuron connections between neurons of a neural network based on the example data pairs.

### Claim 36 (depends on 31)

The method of claim 31, further comprising selecting the denoising autoencoder that is specific to the first database.

### Claim 37 (depends on 36)

The method of claim 36, wherein the denoising autoencoder specific to the first database comprises a neural network that comprises an input layer, a hidden layer, and an output layer, and wherein the hidden layer includes fewer neurons than the input layer.

### Claim 38 (depends on 31)

The method of claim 31, further comprising, providing a data pair comprising the first input data vector and the second data vector as an additional training example data pair to the denoising autoencoder to train the denoising autoencoder to denoise input from the first database.

### Claim 39 (depends on 31)

The method of claim 31, further comprising:
comparing the first input data vector with a plurality of validated data vectors, wherein the comparison includes performing binary pairwise comparisons between elements of the first input data vector and elements of the plurality of validated data vectors; and
calculating a percentage of pairwise comparisons that resulted in a match.

### Claim 40 (depends on 39)

The method of claim 39, further comprising, continuously updating denoising autoencoder each time a match is made between the first input data vector and one or more of the plurality of validated data vectors.

### Claim 41

A system comprising:
communications circuitry of a device configured to receive a corrupted record from a first database; and
control circuitry of the device configured to:
generate a first input data vector based on the corrupted record using a denoising autoencoder, wherein the denoising autoencoder was trained using a plurality of records from the first database;
generate a second data vector based on a validated record from a validated database; and
in response to determining a match based on comparing the first input data vector with the second data vector:
transmit, using the communications circuitry, the validated record to the first database.

### Claim 42 (depends on 41)

The system of claim 41, wherein, in response to determining that a match does not exist, outputting an indication of a match failure.

### Claim 43 (depends on 41)

The system of claim 41, in response to determining the match:
determining that the corrupted record is incomplete;
causing the first database to complete the corrupted record based on the validated record.

### Claim 44 (depends on 41)

The system of claim 41, wherein the control circuitry is further configured to:
use the additional training example data pair as an input to a neural network;
generate an output based on the additional training example data pair as the input; and
evaluate the neural network's performance based on the output.

### Claim 45 (depends on 44)

The system of claim 44, wherein training the denoising autoencoder using the plurality of records from the first database comprises the control circuitry is configured to adjust neuron connections between neurons of a neural network based on the example data pairs.

### Claim 46 (depends on 41)

The system of claim 41, wherein the control circuitry is further configured to select the denoising autoencoder that is specific to the first database.

### Claim 47 (depends on 46)

The system of claim 46, wherein the denoising autoencoder specific to the first database comprises a neural network that comprises an input layer, a hidden layer, and an output layer, and wherein the hidden layer includes fewer neurons than the input layer.

### Claim 48 (depends on 41)

The system of claim 41, wherein the control circuitry is further configured to provide a data pair comprising the first input data vector and the second data vector as an additional training example data pair to the denoising autoencoder to train the denoising autoencoder to denoise input from the first database.

### Claim 49 (depends on 41)

The system of claim 41, wherein the control circuitry is further configured to:
compare the first input data vector with a plurality of validated data vectors, wherein the comparison includes performing binary pairwise comparisons between elements of the first input data vector and elements of the plurality of validated data vectors; and
calculate a percentage of pairwise comparisons that resulted in a match.

### Claim 50 (depends on 49)

The system of claim 49, wherein the control circuitry is further configured to continuously update the denoising autoencoder each time a match is made between the first input data vector and one or more of the plurality of validated data vectors.
