Complementary Item Discovery Using Precomputed Recommendation Graphs

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing search technologies struggle to efficiently identify and provide complementary items, leading to increased computing resource consumption, network latency, and storage device I/O costs due to repetitive user queries and item renderings when dealing with large item listings.

Innovation Solution

A recommendation system leveraging multi-target search to detect objects in images and associate them with complementary item listings, using machine learning to predict and provide item listing recommendations and query suggestions based on user interactions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional recommendation systems recommend similar items, then item similarity is maintained, but complementary item discovery is limited

Engineering Contradiction:

Improvecomplementary item discoveryVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by pre-computing complementary item relationships and storing them in a graph structure during system initialization. This allows the system to have complementary item data ready in advance, eliminating the need for complex real-time computations when users request recommendations, thus resolving the contradiction between comprehensive complementary item discovery and system complexity.

Inventive Principle:

Principle #10Preliminary action

2Productivity

If search engines process multiple user queries to find complementary items, then item recommendations are provided, but computing resource consumption increases

Engineering Contradiction:

Improveitem recommendation efficiencyVSAvoidcomputing resource consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The system pre-computes and stores complementary item relationships in a graph structure during initialization, so that when users request recommendations, the system can quickly retrieve pre-computed results without performing complex real-time searches. This preliminary action dramatically reduces computing resource consumption during actual user interactions while maintaining high recommendation efficiency.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system creates a copied representation of item relationships in the form of a graph structure that mirrors the complementary item associations. This copied graph allows rapid querying and recommendation generation without needing to access and process the entire item database repeatedly, thus reducing computing resource consumption while maintaining productivity.

Inventive Principle:

Principle #26Copying

3Reliability

If search engines perform repetitive queries for complementary items, then comprehensive results are obtained, but network latency increases

Engineering Contradiction:

Improvecomplementary item identification accuracyVSAvoidnetwork latency

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system pre-computes complementary item relationships and stores them in a graph structure during initialization. When users request recommendations, the system retrieves pre-computed results from this graph rather than performing repetitive real-time searches. This preliminary action ensures reliable complementary item identification while dramatically reducing network latency by eliminating repeated query cycles.

Inventive Principle:

Principle #10Preliminary action

4Loss of information

If search engines repeatedly render item results for complementary items, then complete information is provided, but storage device I/O costs increase

Engineering Contradiction:

Improveitem information completenessVSAvoidstorage device I/O costs

Core Design Contradiction:

Loss of informationVSLoss of energy

Solution Approach 1:

The system creates a copied graph structure that contains complementary item relationships, allowing it to retrieve recommendation data from this compact copy rather than repeatedly accessing and rendering full item details from the main database. This copying approach provides complete item information through the graph structure while significantly reducing storage device I/O costs by minimizing repeated data access operations.

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20250225567A1Complementary item recommendation system

Publication Date: 2025.07.10 EBAY INC

AI summary

A recommendation system leverages multi-target search to provide item listing recommendations and/or query suggestions. For a given input image with multiple objects, multi-target search uses object detection to detect each object, and stores complementary object data associating each object from the image. Additionally, a search of an item listing datastore is performed using each object from the image as a search query. Based on item listings returned as search results, complementary item listings data associating item listings is stored. In some configurations, the complementary item listings data is also used to train a machine learning model to predict complementary item listings for a given item listing. When an input item listing is received, item listing recommendations and/or query suggestions are determined for the input item listing using the complementary object data, the complementary item listing data, and/or the machine learning model.