Weakly Supervised Attribute Extraction for Unstructured Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current online concierge systems face challenges in efficiently extracting attributes from unstructured data describing items, due to the diversity of potential attributes and the large number of items, which requires significant computational resources and manual effort for data structuring.

Innovation Solution

The online concierge system employs natural language processing methods to extract candidate values for attributes from item names, generates a seed set of values based on user inputs, and clusters candidate values to reduce the number of attributes users need to review, thereby improving efficiency and accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual review methods are used to extract attributes from unstructured data, then data accuracy is improved, but productivity deteriorates due to resource intensity and difficulty to scale

Engineering Contradiction:

Improveattribute extraction accuracyVSAvoiddata processing throughput

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system enables self-service attribute extraction by training machine learning models to automatically process unstructured data. The models learn from training datasets containing unstructured item descriptions and their corresponding attributes, then autonomously extract attributes from new unstructured data without manual intervention, achieving both high accuracy and scalable productivity

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical manual review process with automated machine learning systems. Instead of human reviewers manually extracting attributes, the system uses trained models that process unstructured data computationally, substituting human cognitive work with automated algorithmic processing that maintains accuracy while dramatically increasing throughput

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Loss of information

If comprehensive attribute extraction is performed on all items, then information completeness is improved, but loss of time increases due to large numbers of items and attribute diversity

Engineering Contradiction:

Improveattribute coverage completenessVSAvoidprocessing time per item

Core Design Contradiction:

Loss of informationVSLoss of time

Solution Approach 1:

The system performs preliminary action by pre-training machine learning models on comprehensive training datasets before deployment. The models learn attribute extraction patterns in advance from diverse examples, enabling them to quickly and accurately extract attributes from new items without requiring time-consuming processing for each individual item

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent applies parameter changes by adjusting model confidence thresholds and filtering parameters to balance completeness and speed. The system can dynamically adjust extraction sensitivity based on item characteristics, processing high-confidence extractions quickly while applying more thorough analysis only when needed, thereby reducing overall processing time while maintaining information completeness

Inventive Principle:

Principle #35Parameter changes

3Manufacturing precision

If multiple machine learning models are trained for different attribute categories, then manufacturing precision is improved, but device complexity increases

Engineering Contradiction:

Improveattribute classification accuracyVSAvoidmodel architecture complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies universality by designing a unified machine learning model architecture that can handle multiple attribute categories simultaneously. Instead of separate models for each attribute type, the system uses a single model trained on diverse training data covering multiple categories, enabling it to extract different attribute types through the same computational pathway, thereby reducing complexity while maintaining precision

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system uses segmentation by dividing the training process into structured phases where different attribute types are learned in organized sequences. The model processes unstructured data through segmented layers that specialize in different attribute extraction tasks, allowing complex multi-category classification to be achieved through modular, manageable components rather than a monolithic complex system

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250117442A1Weakly supervised extraction of attributes from unstructured data to generate training data for machine learning models

Publication Date: 2025.04.10 MAPLEBEAR INC

AI summary

An online concierge system receives unstructured data describing items offered for purchase by various warehouses. To generate attributes for products from the unstructured data, the online concierge system extracts candidate values for attributes from the unstructured data through natural language processing. One or more users associate a subset candidate values with corresponding attributes, and the online concierge system clusters the remaining candidate values with the candidate values of the subset associated with attributes. One or more users provide input on the accuracy of the generated clusters. The candidate values are applied as labels to items by the online concierge system, which uses the labeled items as training data for an attribute extraction model to predict values for one or more attributes from unstructured data about an item.