# US 2021/0174032 A1 · Compliance graph generation

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

- Number: US 2021/0174032 A1
- Document ID: US20210174032A1
- Status: Published application
- Kind: A1
- Application number: 16/704,801
- Filed: 2019-12-05
- Published: 2021-06-10
- Earliest claimed: 2019-12-05
- Assignees: Intuit Inc.
- Applicants: Intuit Inc.
- Inventors: Conrad DE PEUTER
- Web page: https://patentgrep.com/patent/US20210174032A1

## CPC

- G06F 40/30: Semantic analysis
- G06F 40/45: Example-based machine translation; Alignment
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 5/02: Knowledge representation; Symbolic representation
- G06N 20/00: Machine learning
- G06N 3/02: Neural networks
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Certain aspects of the present disclosure provide techniques for generating a compliance graph based on a compliance rule to implement in a software program product for determining user compliance. To generate a compliance graph, an encoder receives a compliance rule in a source language and generates a set of corresponding vectors. The decoder, which has been trained using verified training pairs and synthetic data, generates a sequence of operations based on the vectors from the encoder. The sequence of operations is the used to build a graph in which each operation is a node in the graph and each node is connected to at least one other node in the same graph or a separate graph.

## Claims

### Claim 1

A method, comprising:
receiving one or more compliance rules in a source language;
generating, via a pre-trained Universal Language Model (ULM) encoder, a set of vectors corresponding to each compliance rule;
generating, via a trained decoder and based on the set of vectors, a sequence in a target language corresponding to each compliance rule, the sequence comprising a set of operations; and
generating a compliance graph of each sequence in the target language by:
identifying each operation in the sequence;
splitting the sequence based on each identified operation; and
connecting each operation with at least one other operation.

### Claim 2 (depends on 1)

The method of claim 1, wherein the decoder is trained with:
a set of verified training pairs of a type of source language and target language; and
a set of synthetic data generated based on the set of verified training pairs.

### Claim 3 (depends on 2)

The method of claim 2, wherein the method further comprises:
receiving verification of a training pair from a user;
determining a confidence level associated with the verified training pair; and
generating synthetic data with the same confidence level.

### Claim 4 (depends on 2)

The method of claim 2, wherein the source language of the compliance rule and the target language of the sequence is the same as the type of source language and target language that trained the decoder.

### Claim 5 (depends on 2)

The method of claim 2, wherein the source language of the compliance rule and the target language of the sequence is different than the type of source language and target language that trained the decoder.

### Claim 6 (depends on 2)

The method of claim 2, wherein the set of verified training pairs includes more than one type of source language and target language.

### Claim 7 (depends on 1)

The method of claim 1, wherein a compliance graph with two or more operations is generated for a sequence by linking one operation to at least one other operation based on a reference to the other operation.

### Claim 8 (depends on 1)

The method of claim 1, wherein a compliance graph with one operation is generated for a sequence and is connected to another compliance graph based on a reference to the one operation.

### Claim 9 (depends on 1)

The method of claim 1, wherein each generated compliance graph is connected to at least one other generated compliance graph based on a reference in one compliance graph to another operation in another compliance graph.

### Claim 10 (depends on 1)

The method of claim 1, wherein each operation in the sequence is a node in the compliance graph.

### Claim 11

A system, comprising:
a processor; and
a memory storing instructions which when executed by the processor perform a method comprising:
receiving one or more compliance rules in a source language;
generating, via a pre-trained Universal Language Model (ULM) encoder, a set of vectors corresponding to each compliance rule;
generating, via a trained decoder and based on the set of vectors, a sequence in a target language corresponding to each compliance rule, the sequence comprising a set of operations; and
generating a compliance graph of each sequence in the target language by:
identifying each operation in the sequence;
splitting the sequence based on each identified operation; and
connecting each operation with at least one other operation.

### Claim 12 (depends on 11)

The system of claim 11, wherein the decoder is trained with:
a set of verified training pairs of a type of source language and target language; and
a set of synthetic data generated based on the set of verified training pairs.

### Claim 13 (depends on 12)

The system of claim 12, wherein the method further comprises:
receiving verification of a training pair from a user;
determining a confidence level associated with the verified training pair; and
generating synthetic data with the same confidence level.

### Claim 14 (depends on 12)

The system of claim 12, wherein the source language of the compliance rule and the target language of the sequence is the same as the type of source language and target language that trained the decoder.

### Claim 15 (depends on 12)

The system of claim 12, wherein the source language of the compliance rule and the target language of the sequence is different than the type of source language and target language that trained the decoder.

### Claim 16 (depends on 11)

The system of claim 11, wherein a compliance graph with two or more operations is generated for a sequence by linking one operation to at least one other operation based on a reference to the other operation.

### Claim 17 (depends on 11)

The system of claim 11, wherein a compliance graph with one operation is generated for a sequence and is connected to another compliance graph based on a reference to the one operation.

### Claim 18 (depends on 11)

The system of claim 11, wherein each generated compliance graph is connected to at least one other generated compliance graph based on a reference in one compliance graph to another operation in another compliance graph.

### Claim 19

A method, comprising:
receiving one or more compliance rules in a source language;
generating, via a pre-trained Universal Language Model (ULM) encoder, a set of vectors corresponding to each compliance rule;
generating, via a trained decoder and based on the set of vectors, a sequence in a target language corresponding to each compliance rule, the sequence comprising a set of operations, wherein the decoder is trained with:
a set of verified training pairs of a type of source language and target language; and
a set of synthetic data generated based on the set of verified training pairs; and
generating a compliance graph of each sequence in the target language by:
identifying each operation in the sequence;
splitting the sequence based on each identified operation; and
connecting each operation with at least one other operation.

### Claim 20 (depends on 19)

The method of claim 19, wherein the method further comprises implementing the compliance graph with a completeness graph in a knowledge engine.
