# US 2019/0171655 A1 · Mtransaction processing improvements

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

- Number: US 2019/0171655 A1
- Document ID: US20190171655A1
- Status: Published application
- Kind: A1
- Application number: 16/209,887
- Filed: 2018-12-04
- Published: 2019-06-06
- Earliest claimed: 2017-12-04
- Assignees: Panjiva, Inc.
- Applicants: Panjiva, Inc.
- Inventors: James Ryan Psota; Jason Seth Prentice; Peter Goodings Swartz; Timothy George Garnett; Giulio Gueltrini
- Web page: https://patentgrep.com/patent/US20190171655A1

## CPC

- G06F 16/288: {Entity relationship models}
- G06F 16/215: Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
- G06F 16/2379: {Updates performed during online database operations; commit processing}
- G06F 16/24578: {using ranking}
- G06F 16/285: {Clustering or classification}
- G06Q 10/083: Shipping
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

The technology features a system and computer-implemented method for resolving a relationship between objects. A target object index is generated based on a group of target objects. One or more lookup operations is performed on each target object in the target object index for each source object in a group of source objects. A plurality of source target object pairs is generated, each source target object pair comprising one source object and one target object having at least one matching data value. Each source target object pair is converted into a numeric feature vector. The numeric feature vector is classified corresponding to each source target object pair using a binary classifier. A match score to each source target object pair is applied based on the classification using the binary classifier. Any source target object pair having a match score lower than a match threshold value is discarded.

## Claims

### Claim 1

A computer-implemented method for resolving a relationship between objects, the method comprising:
generating a target object index based on a group of target objects;
performing, for each source object in a group of source objects, one or more lookup operations on each target object in the target object index;
generating a plurality of source target object pairs, each source target object pair comprising one source object and one target object having at least one matching data value;
converting each source target object pair into a numeric feature vector;
classifying the numeric feature vector corresponding to each source target object pair using a binary classifier;
applying a match score to each source target object pair based on the classification using the binary classifier; and
discarding any source target object pair having a match score lower than a match threshold value.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1 further comprising:
ranking the source target object pairs having a match score higher than the match threshold value based on match scores;
selecting a subset of the ranked source target object pairs based on a likelihood that the source object and the target object of each source target object pair of the subset are associated with the same object; and
creating a persistent entry for each source target object pair of the subset, each persistent entry indicating the likelihood that the source object and the target object are associated with the same object.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1 further comprising:
populating a disjoint set data structure with a plurality of entity objects;
ingesting source target object pairs having a match score higher than the match threshold value; and
updating the disjoint set data structure using a union-find algorithm on each source target object pair having a match score higher than the match threshold value.

### Claim 4 (depends on 3)

The computer-implemented method of claim 3 further comprising
generating an edge weighting value for each source target object pair having a match score higher than the match threshold value based on the match score applied to the source target object pair;
determining a set of candidate cuts based on the edge weighting values, each candidate cut relating to one or more edges linking elements of the disjoint set data structure;
scoring each candidate cut based on an expected error rate of the binary classifier;
performing one of the candidate cuts based on a score assigned by the binary classifier being above a cut threshold value.

### Claim 5 (depends on 4)

The computer-implemented method of claim 4 wherein performing one of the candidate cuts further comprises:
creating a plurality of sub-clusters, each sub-cluster comprising a subset of elements of the disjoint set data structure.

### Claim 6 (depends on 5)

The computer-implemented method of claim 5 further comprising:
recursively repeating the determining, scoring, and performing steps for each sub-cluster of the plurality of sub-clusters; and
defining a new entity object for each sub-cluster based on a score assigned by the binary classifier being below the cut threshold value.

### Claim 7 (depends on 6)

The computer-implemented method of claim 6 wherein a plurality of server computing devices perform the recursively repeating step for the plurality of sub-clusters, each sub-cluster of the plurality of sub-clusters being operated on by a different server computing device substantially in parallel.

### Claim 8 (depends on 4)

The computer-implemented method of claim 4 wherein determining a set of candidate cuts further comprises applying a candidate partition algorithm to the disjoint set data structure.

### Claim 9 (depends on 8)

The computer-implemented method of claim 8 wherein the candidate partition algorithm is at least one of Karger's algorithm, the Karger-Stein algorithm, and a spectral clustering algorithm.

### Claim 10 (depends on 1)

The computer-implemented method of claim 1 wherein each target object in the group of target objects comprises:
a target object identifier; and
at least one related value.

### Claim 11 (depends on 10)

The computer-implemented method of claim 10 wherein each target object further comprises a target object partition key.

### Claim 12 (depends on 10)

The computer-implemented method of claim 10 wherein the at least one related value matches a value of at least one other target object in the group of target objects.

### Claim 13 (depends on 10)

The computer-implemented method of claim 10 further comprising performing a cleaning operation on the at least one related value of each target object.

### Claim 14 (depends on 13)

The computer-implemented method of claim 13 wherein performing the cleaning operation further comprises one or more of normalizing the case of text, normalizing punctuation of text, and expanding abbreviations of text.

### Claim 15 (depends on 13)

The computer-implemented method of claim 13 wherein the cleaning operation is performed prior to generating the index.

### Claim 16 (depends on 11)

The computer-implemented method of claim 11 further comprising generating a separate index for each value of the partition key.

### Claim 17 (depends on 1)

The computer-implemented method of claim 1 wherein each source object in the group of source objects comprises a source object identifier.

### Claim 18 (depends on 17)

The computer-implemented method of claim 17 wherein each source object in the group of source objects further comprises a source object partition key.

### Claim 19 (depends on 1)

The computer-implemented method of claim 1 wherein the binary classifier comprises a pretrained machine learning model based on one or more of a random forest method, logistic regression, a support vector machine, and a neural network.

### Claim 20 (depends on 1)

The computer-implemented method of claim 1 further comprising performing a cleaning operation on one or more data values of each source object of a group of source objects.
