EP-4738229-A1 - AUTOMATED PROCESSING USING MACHINE LEARNING GENERATED GRAPH-BASED RULES

EP4738229A1EP 4738229 A1EP4738229 A1EP 4738229A1EP-4738229-A1

Abstract

Technical solutions are directed to automating processing using machine learning and graph structure based rules. A processor can identify a graph data structure that connects, using semantic edges, a plurality of components in accordance with a taxonomy. The processor can detect, using the graph data structure, a change to a protocol used to perform an operation comprising one or more electronic transactions between electronic accounts related to the plurality of components. The processor can generate, using the graph data structure, responsive to detection of the change, one or more rules to perform the operation in accordance with the change to the protocol. The processor can construct a prompt with the one or more rules and at least a portion of an electronic document. The processor can execute, using a model trained with a generative machine learning technique, the operation based on the prompt.

Inventors

SILVA, Allan Barcelos
WESTHELLE, Matheus
TOUNSI, Ash

Assignees

ADP, Inc.

Dates

Publication Date: 20260506
Application Date: 20251031

Claims (15)

A system comprising: one or more processors, coupled with memory, to: identify a knowledge graph data structure that connects, using semantic edges, a plurality of components in accordance with a taxonomy; detect, using the knowledge graph data structure, a change to a protocol used to perform an operation comprising one or more electronic transactions between electronic accounts related to the plurality of components; generate, using the knowledge graph data structure, responsive to detection of the change, one or more rules to perform the operation in accordance with the change to the protocol; construct a prompt with the one or more rules and at least a portion of an electronic document; and execute, using a model trained with a generative machine learning technique, the operation based on the prompt.
The system of claim 1, wherein the one or more processors are further configured to: identify a dataset comprising documents for electronic transactions for a plurality of payroll operations; determine, using the dataset, the components corresponding to electronic transactions for each of the plurality of payroll operations; determine, using the dataset, the semantic edges defining relationships between the components; and generate, using the components and the semantic edges, the knowledge graph data structure for the plurality of payroll operations.
The system of claim 1 or 2, wherein the one or more processors are further configured to: identify metadata corresponding to an electronic account for which the operation is to be executed; identify, based on the metadata, one or more components from the plurality of components that correspond to the operation for the electronic account; and construct the prompt, using the metadata and the one or more components, for the operation to be executed.
The system of any preceding claim, wherein the plurality of components includes portions of documents corresponding to processing of taxes for an electronic account of the electronic accounts, the electronic account corresponding to at least one of an enterprise or an employee of the enterprise; and/or wherein the semantic edges include at least one of: a hierarchical semantic edge determined based on a hierarchy of two or more components, a causal semantic edge representing a cause-and-effect relationship between two or more components, or a temporal semantic edge indicative of a timing of events corresponding to two or more components; and/or wherein the knowledge graph data structure is represented in a JavaScript Object Notation (JSON) format and the components and the semantic edges are represented using JSON objects.
The system of any preceding claim, wherein the one or more processors are further configured to: receive a document including updated regulations impacting the electronic transactions; and identify, based on the document, the change to the protocol; and update the knowledge graph data structure based on the identified change.
The system of any preceding claim, wherein the one or more processors are further configured to: generate a user interface for interacting with the knowledge graph data structure; receive, via the user interface, an adjustment to at least one of a component of the plurality of components or a semantic edge of the semantic edges defining relationships between the plurality of components; update, based on the adjustment, the knowledge graph data structure; and display the updated knowledge graph data structure via the user interface.
The system of any preceding claim, wherein the one or more processors are further configured to: identify one or more machine learning (ML) models trained on a dataset of a plurality of documents for performing a plurality of electronic transactions according to a plurality of protocols; generate, using the one or more ML models, the knowledge graph data structure; and optionally wherein the one or more processors are further configured to: identify a document for performing one or more electronic transactions that is not included in the plurality of documents; update, based on the document input into the one or more ML models, the knowledge graph data structure.
The system of any preceding claim, wherein the one or more processors are further configured to: identify a geographical area corresponding to an electronic account of the electronic accounts; and select the protocol responsive to a match of a geographical data of the protocol with the geographical area corresponding to the electronic account; and optionally wherein the knowledge graph data structure includes components and semantic edges corresponding to a plurality of protocols comprising the protocol, the plurality of protocols for operations corresponding to a plurality of geographical areas comprising the geographical area.
A method, comprising identifying, by one or more processors coupled with memory, a knowledge graph data structure that connects, using semantic edges, a plurality of components in accordance with a taxonomy; detecting, by the one or more processors, using the knowledge graph data structure, a change to a protocol used to perform an operation comprising one or more electronic transactions between electronic accounts related to the plurality of components; generating, by the one or more processors, using the knowledge graph data structure, responsive to detection of the change, one or more rules to perform the operation in accordance with the change to the protocol; constructing, by the one or more processors, a prompt with the one or more rules and at least a portion of an electronic document; and causing, by the one or more processors, using a model trained with a generative machine learning technique, execution of the operation based on the prompt.
The method of claim 9, comprising: identifying, by the one or more processors, a dataset comprising documents for electronic transactions for a plurality of payroll operations; determining, by the one or more processors, using the dataset, the components corresponding to electronic transactions for each of the plurality of payroll operations; determining, by the one or more processors, using the dataset, the semantic edges defining relationships between the components; and generating, by the one or more processors, using the components and the semantic edges, the knowledge graph data structure for the plurality of payroll operations.
The method of claim 9 or 10, comprising: identifying, by the one or more processors, metadata corresponding to an electronic account for which the operation is to be executed; identifying, by the one or more processors, based on the metadata, one or more components from the plurality of components that correspond to the operation for the electronic account; and constructing, by the one or more processors, the prompt, using the metadata and the one or more components, for the operation to be executed.
The method of any of claims 9 to 11, comprising: receiving, by the one or more processors, a document including updated regulations impacting the electronic transactions; and identifying, by the one or more processors, based on the document, the change to the protocol; and updating, by the one or more processors, the knowledge graph data structure based on the identified change.
The method of any of claims 9 to 12, comprising: generating, by the one or more processors, a user interface for interacting with the knowledge graph data structure; receiving, by the one or more processors, via the user interface, an adjustment to at least one of a component of the plurality of components or a semantic edge of the semantic edges defining relationships between the plurality of components; updating, by the one or more processors, based on the adjustment, the knowledge graph data structure; and displaying, by the one or more processors, the updated knowledge graph data structure via the user interface.
The method of any of claims 9 to 13, comprising: identifying, by the one or more processors, one or more machine learning (ML) models trained on a dataset of a plurality of documents for performing a plurality of electronic transactions according to a plurality of protocols; generating, by the one or more processors, using the one or more ML models, the knowledge graph data structure; and/or. the method of comprising: identifying, by the one or more processors, a document for performing one or more electronic transactions that is not included in the plurality of documents; updating, by the one or more processors, based on the document input into the one or more ML models, the knowledge graph data structure.
A non-transient computer readable medium comprising processor readable instructions which, when executed by one or more processors, cause the one or more processors to carry out the method of any of claims 9 to 14.

Description

TECHNICAL FIELD This application is generally related to computing technology, and particularly to a computing technology solution for automated payroll processing using machine learning. BACKGROUND Data processing technologies can automatically make decisions, provide predictive analytics, and streamline data management. However, as the data relationships and transactional processes within digital ecosystems become increasingly intricate, it can be challenging for data processing systems to effectively, efficiently, and reliably navigate such interdependencies while accurately and consistently making automatic decisions and predicting analytics. SUMMARY The technical solutions described herein automate processing operations using a machine learning (ML)-generated graph structure to navigate the complexities of diverse and multi-jurisdictional tax documentation, utilizing rules derived from the graph structure taxonomy to process electronic transactions within payroll systems. For example, the technical solutions can improve the accuracy and reliability with which data processing systems can process complex interrelationships. Thus, the technical solutions facilitate the data processing systems (computing systems) to operate in an efficient manner, thereby improving the accuracy and reliability with which the data processing systems can automatically make decisions and predict analytics. To do so and recognizing that hardcoding changes is impractical and resource-intensive, the technical solutions establish a payroll taxonomy ontology that formalizes relationships between components such as tax documentation forms and computation steps. By employing large language models (LLMs), the graph structure facilitates the automatic generation of rules for processing electronic transactions according to payroll protocols, thereby enhancing compliance and accuracy. This technology significantly improves the effectiveness, availability, and energy efficiency of processing (e.g., payroll processing) by automating the interpretation of tax documents and calculating tax payments or deductions based on client account data. An aspect of the technical solutions can be directed to a system. The system can include one or more processors coupled with memory. The one or more processors can identify a knowledge graph data structure that connects, using semantic edges, a plurality of components in accordance with a taxonomy. The one or more processors can detect, using the knowledge graph data structure, a change to a protocol used to perform an operation comprising one or more electronic transactions between electronic accounts related to the plurality of components. The one or more processors can generate, using the knowledge graph data structure, responsive to detection of the change, one or more rules to perform the operation in accordance with the change to the protocol. The one or more processors can construct a prompt with the one or more rules and at least a portion of an electronic document. The one or more processors can execute, using a model trained with a generative machine learning technique, the operation based on the prompt. The one or more processors can be configured to identify a dataset comprising documents for electronic transactions for a plurality of payroll processes. The one or more processors can be configured to determine, using the dataset, the components corresponding to electronic transactions for each of the plurality of payroll processes. The one or more processors can be configured to determine, using the dataset, the semantic edges defining relationships between the components. The one or more processors can be configured to generate, using the components and the semantic edges, the knowledge graph data structure for the plurality of payroll processes. The one or more processors can be configured to identify metadata corresponding to an electronic account for which the operation is to be executed. The one or more processors can be configured to identify, based on the metadata, one or more components from the plurality of components that correspond to the operation for the electronic account. The one or more processors can be configured to construct the prompt, using the metadata and the one or more components, for the operation to be executed. The plurality of components can include portions of documents corresponding to processing of taxes for an electronic account of the electronic accounts. The electronic account can correspond to at least one of an enterprise or an employee of the enterprise. The semantic edges can include at least one of: a hierarchical semantic edge determined based on a hierarchy of two or more components, a causal semantic edge representing a cause-and-effect relationship between two or more components, or a temporal semantic edge indicative of a timing of events corresponding to two or more components. The knowledge graph data structure can be represented in a JavaScript