Machine Learned Search Recommendations from Annotated Catalog Text
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional search and recommendation systems in e-commerce struggle with high computational costs when converting unstructured data into structured data, and product knowledge graphs face challenges in accessing complex schemas and developing general-purpose algorithms for various product types.
Innovation Solution
An online concierge system uses machine learned models to convert structured data into annotated text data, generating templates and prompts for AI systems to train search and recommendation models, allowing for efficient processing of unstructured data and improving recommendation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If unstructured data is converted to structured data through information extraction, then structured data can be used in conventional search and recommendation systems, but computational costs become extremely high
Solution Approach 1:
Instead of converting unstructured data to structured data as in conventional systems, this patent inverts the approach by training machine learned models directly on unstructured data (text, images, videos). The models learn patterns and representations from the raw unstructured data without requiring it to be converted into structured formats, thereby eliminating the high computational cost of information extraction while maintaining data usability.
2Loss of information
If product knowledge graphs use complex schemas or ontologies to model knowledge, then comprehensive product knowledge can be captured, but accessing such knowledge becomes challenging
Solution Approach 1:
The patent replaces the mechanical system of complex schemas and ontologies with machine learned models that automatically learn knowledge representations from unstructured data. Instead of manually designing complex knowledge graph schemas, the models learn patterns, relationships, and knowledge structures directly from the data, making knowledge accessible through model predictions rather than navigating complex schemas.
3Adaptability or versatility
If general-purpose algorithms are developed to convert unstructured data to structured format, then any product type can be processed, but each product type requires algorithms specific to that product type
Solution Approach 1:
The patent creates a universal machine learned model that can process any product type without requiring product-specific algorithms. The model is trained on diverse unstructured data from multiple product categories and learns general patterns that apply across different product types. This single multi-functional model replaces the need for developing separate algorithms for each product type, achieving versatility without proportional increases in complexity.
Data Source
AI summary
A system may generate a prompt based in part on a search query from a customer client device. The prompt instructs a machine learned model to provide item predictions. And the model was trained by: converting structured data describing items of an online catalog to annotated text data (unstructured data), generating training examples based in part on the annotated text data, and training the model using the training examples. The system may receive item predictions generated by the prompt being applied to the machine learned model, the item predictions may have corresponding item identifiers. The item predictions are processed to identify a recommended item from the item predictions. The processing includes determining item information for the recommended item using an item identifier associated with the recommended item. The item information is provided to the customer client device.