GAN Generator for Query-Key Term Matching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing computer platforms for delivering digital content items are inefficient in matching user queries to key terms, leading to slow content delivery and suboptimal resource utilization, particularly with semantic-based matching algorithms.

Innovation Solution

A computer-implemented technique using a generator component based on a sequence-to-sequence neural network trained with an adversarial generative network (GAN) system, which generates key terms without requiring comparison to actual bid key terms or historical query-key-term pairs, enabling fast and resource-efficient content delivery.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If semantic-based matching algorithms are used to match queries to key terms, then matching accuracy is improved, but system resource consumption increases and content delivery speed decreases

Engineering Contradiction:

Improvematching accuracyVSAvoidcontent delivery speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system pre-trains the generator component offline using大量 query-key term pairs to learn the mapping relationship between queries and key terms. This preliminary training allows the generator to rapidly generate accurate key terms during online operation without requiring resource-intensive semantic matching algorithms at query time, thus resolving the contradiction between matching accuracy and content delivery speed

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional semantic-based matching algorithms with a neural network-based generator component. The generator uses learned patterns from training data to directly generate key terms from queries, substituting the computationally expensive semantic analysis process with a more efficient neural network inference process that maintains accuracy while improving speed and reducing resource consumption

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If traditional query-key term matching methods are used, then system resource utilization is improved, but matching quality and relevance decrease

Engineering Contradiction:

Improveresource utilization efficiencyVSAvoidmatching quality

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent replaces traditional lexical-based and semantic-based matching algorithms with a neural network generator that learns complex query-key term relationships from training data. This substitution enables the system to achieve superior matching quality by capturing nuanced patterns that traditional algorithms miss, while the generator's efficient inference process maintains good resource utilization

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the fundamental approach from rule-based and statistical matching to neural network-based generative matching. By transforming the matching task into a key term generation task supervised by a discriminator, the system achieves higher matching quality through learned representations while maintaining computational efficiency through the generator's optimized inference process

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If comparison with actual bid key terms and historical log mining are performed, then key term matching accuracy is improved, but processing time and computational resources increase

Engineering Contradiction:

Improvekey term matching accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs comprehensive learning from historical query-key term pairs during offline training, incorporating patterns from actual bid key terms and historical logs into the generator's knowledge base. This preliminary action transfers the computational burden to the training phase, allowing rapid online inference that generates accurate key terms without real-time comparison or log mining

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The generator component learns to replicate the matching behavior of traditional systems by training on historical query-key term pairs and bid key terms. Instead of copying and comparing actual key terms at query time, the generator creates synthesized key terms that capture the essential patterns from training data, achieving similar accuracy with much faster processing

Inventive Principle:

Principle #26Copying

Data Source

PatentUS10642846B2Using a generative adversarial network for query-keyword matching

Publication Date: 2020.05.05 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

A computer-implemented technique is described herein for providing a digital content item using a generator component. The generator component corresponds to a sequence-to-sequence neural network that is trained using an adversarial generative network (GAN) system. In one approach, the technique involves: receiving a query from a user computing device over a computer network; generating random information; generating a key term using the generator component based on the query and the random information; selecting at least one content item based on the key term; and sending the content item(s) over the computer network to the user computing device.