Unified Attribute Extraction Model for E-Commerce Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current systems fail to extract both closed and open list attributes from text-based descriptions of products in a unified manner, requiring separate configurations and implementations for each attribute type.

Innovation Solution

A system utilizing an extended conditional random field (XCRF) process to extract closed list attributes from text-based descriptions, enabling a unified attribute extraction model that handles both closed and open list attributes using a shared set of latent variables in a hidden layer.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If separate attribute extraction systems are used for closed list and open list attributes, then each attribute type can be extracted with dedicated configuration, but the system complexity increases and requires multiple implementations

Engineering Contradiction:

Improveattribute extraction accuracyVSAvoidsystem configuration complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent combines separate attribute extraction systems for closed list and open list attributes into a single unified attribute extraction system. The system uses a single neural network model that can handle both attribute types simultaneously, eliminating the need for separate configurations and implementations while maintaining extraction accuracy through a unified architecture that processes both attribute types through shared layers.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The unified attribute extraction system is designed with multi-functionality to handle both closed list and open list attributes through a single model. The system uses a universal architecture with shared embedding layers and hidden layers that can process different attribute types, making the system adaptable to various attribute extraction tasks without requiring separate specialized systems.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If multiple separate attribute extraction systems are implemented, then comprehensive attribute coverage is achieved, but computational resources and training data requirements increase

Engineering Contradiction:

Improveattribute type coverageVSAvoidcomputational resource consumption

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent merges multiple attribute extraction systems into one unified system that processes both closed list and open list attributes simultaneously. This consolidation reduces computational overhead by eliminating redundant processing pipelines and shared infrastructure, while the unified model architecture allows efficient resource utilization through shared computational components.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The unified attribute extraction system provides universal coverage for both closed list and open list attributes through a single multi-functional model. The system uses shared embedding layers and hidden layers that can process different attribute types, reducing the need for separate training data sets and computational resources while maintaining comprehensive attribute coverage.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Reliability

If separate configurations are used for different attribute types, then each attribute can be optimized independently, but the overall system requires more maintenance and updates

Engineering Contradiction:

Improveattribute extraction reliabilityVSAvoidsystem maintenance ease

Core Design Contradiction:

ReliabilityVSEase of operation

Solution Approach 1:

The patent combines separate attribute extraction configurations into a unified system where both closed list and open list attributes are processed through a single model. This unification simplifies maintenance and updates by eliminating the need to manage multiple separate systems, while the modular architecture allows independent optimization of specific attribute processing components within the unified framework.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The unified attribute extraction system provides universal processing capability for different attribute types through a single configurable model. The system maintains reliability by using shared processing components that can be updated and maintained centrally, while still allowing specific optimizations for different attribute types through configurable parameters and loss functions within the unified architecture.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11645318B2Systems and methods for unified extraction of attributes

Publication Date: 2023.05.09 WALMART APOLLO LLC

AI summary

Systems and methods for unified attribute extraction are disclosed. A set of product data including at least one text-based description of each of the products in the set of products is received and at least one closed list attribute is extracted from the at least one text-based description of each of the products. The at least one closed list attribute is extracted by an attribute extraction model configured to implement an extended conditional random field (XCRF) process. A set of attributes including each of the at least one closed list attributes extracted from the at least one text-based description of each of the products in the set of products is output.