Generative Model Output for Semantic Item Retrieval

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing item retrieval systems on listing platforms face inefficiencies in search and recommendation technologies, leading to increased computing resource consumption, repetitive user queries, and irrelevant item listings, due to limited query understanding, insufficient filtering, and unstructured data handling.

Innovation Solution

An item retrieval system leveraging generative model output to generate text for item listings, using query and item embeddings, and a key-value store to reduce latency and resource consumption by providing relevant item listings without repetitive user inputs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional search and recommendation systems are used to retrieve items from a large number of listings, then item retrieval functionality is provided, but computing resource consumption increases and latency is high

Engineering Contradiction:
Improveitem retrieval efficiencyVSAvoidcomputing resource consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The system pre-generates embeddings for queries and items offline before actual retrieval operations. These pre-computed embeddings are stored in embedding stores, allowing the system to avoid expensive real-time embedding computations during user queries, thus reducing latency and computing resource consumption while maintaining retrieval efficiency

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces embeddings as an intermediary representation layer between queries and items. Instead of directly comparing raw queries with item listings, the system transforms both into embedding vectors that capture semantic meaning. This intermediary representation enables efficient similarity computation and reduces the computational burden of traditional text matching methods

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If traditional search systems operate on keyword-based queries, then simple search functionality is provided, but query understanding is limited

Engineering Contradiction:
Improvequery understanding capabilityVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system replaces traditional keyword-based mechanical search with semantic embedding-based retrieval. Instead of matching exact keywords or using simple inverted indexes, the system uses neural network-generated embeddings that capture semantic relationships, enabling the system to understand paraphrases, synonyms, and contextual meanings of queries

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If recommendation systems provide diverse recommended items, then item variety is increased, but relevance to user intent decreases

Engineering Contradiction:
Improveitem relevanceVSAvoiditem diversity
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system uses user interaction feedback (clicks, purchases, views) to refine and update item embeddings and recommendation strategies. This feedback loop allows the system to learn from user behavior and improve the relevance of recommended items over time while maintaining diversity through exploration-exploitation balances in the recommendation algorithm

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4614356A1Item retrieval using generative model output
Publication Date: 2025.09.10 EBAY INC
  • EP4614356A1 patent drawingFigure 1
  • EP4614356A1 patent drawingFigure 2
  • EP4614356A1 patent drawingFigure 3

AI summary

Some aspects relate to technologies for leveraging model output from a generative model to perform item retrieval on a listing platform. In some examples, input for item retrieval is provided to a generative model to produce a model output. A lookup is performed on a key-value store using the model output. If a matching query is found in the key-value store, a query embedding corresponding to the matching query is returned. If a matching query is not found, a query embedding is obtained by generating the query embedding from the model output or using the model output to query a known query index for a known query, which is used to lookup a query embedding in the key-value store. One or more item embeddings are identified based on the query embedding. An output is provided identifying one or more item listings corresponding to the one or more item embeddings.