Language-Model Graph Building for Node Deduplication

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Challenges arise in building effective graphs from disparate and non-uniform data sources due to varying identifiers and descriptors, leading to duplication and inefficiencies in data storage and processing.

Innovation Solution

Utilizing language models to determine object taxonomies and attributes from textual descriptions, enabling standardized graph construction by embedding and similarity scoring to deduplicate and connect nodes based on shared attributes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If traditional graph building methods are used with disparate data sources, then graph construction can be performed, but data duplication and storage inefficiencies occur due to varying identifiers and descriptors

Engineering Contradiction:

Improvestorage requirementsVSAvoiddata standardization

Core Design Contradiction:

Quantity of substanceVSReliability

Solution Approach 1:

The patent transforms unstandardized textual descriptors into standardized structured data by changing the parameter representation format. Language models convert free-text attributes into consistent schemas with fixed data types and formats, enabling reliable comparison and deduplication while reducing storage requirements through efficient data representation.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces language models as an intermediary layer between disparate data sources and the graph construction process. This intermediary standardizes varying identifiers and descriptors from multiple sources into a unified format, enabling reliable graph building without data duplication while maintaining data integrity from diverse sources.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Productivity

If manual graph construction methods are used, then graph building can be performed, but processing time and computational resources increase significantly

Engineering Contradiction:

Improvegraph building efficiencyVSAvoidprocessing time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent replaces manual mechanical graph construction processes with automated language model-based processing. The system automatically extracts entities, attributes, and relationships from unstandardized text, eliminating time-consuming manual data cleaning and standardization steps while significantly improving graph building productivity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary data standardization and entity extraction using language models before graph construction begins. By pre-processing unstandardized textual data into structured formats with consistent schemas, the system eliminates subsequent processing delays and accelerates the overall graph building workflow.

Inventive Principle:

Principle #10Preliminary action

3Adaptability or versatility

If unstandardized data sources are used for graph construction, then more data can be incorporated, but data quality and consistency deteriorate due to varying identifiers and descriptors

Engineering Contradiction:

Improvedata source compatibilityVSAvoiddata consistency

Core Design Contradiction:

Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent creates a universal standardized schema that can accommodate data from multiple diverse sources. The language models map various source-specific identifiers and descriptors to a common standardized format, enabling the system to ingest data from heterogeneous sources while maintaining consistent data quality and precision throughout the graph.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20260010568A1Graph building using language models

Publication Date: 2026.01.08 PAYPAL INC

AI summary

The disclosed computer-implemented method may include determining a taxonomy of an object from its textual description and also standardized attributes of the object from the description and the taxonomy using a language model, according to embodiments. The method may also include building a graph data structure by using the standardized attributes for a node and connecting the node to other nodes using edges for common attributes. Various other methods, systems, and computer-readable media are also disclosed.