Search Query Classifier Using Co-occurrence Analysis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Internet search engines face challenges in identifying and limiting exposure to sensitive or offensive content, such as child pornography and terrorist recruitment material, as users may inadvertently encounter such content during searches.

Innovation Solution

A method and system for training a classifier to identify search queries seeking inappropriate content by generating a database of known queries, expanding it with user-submitted queries, and using co-occurrence analysis to determine candidate queries associated with sensitive or offensive content, which are then used to modify search results and train the classifier.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a large database of search queries seeking inappropriate content is manually compiled, then the classifier training data quality is improved, but the time and human resources required are excessive

Engineering Contradiction:

Improveclassifier training data qualityVSAvoidtime and human resources for database compilation

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system automatically compiles and expands the training database using its own search query data and co-occurrence analysis algorithms, eliminating the need for extensive manual curation. The classifier trains on self-generated training data, achieving high quality without excessive human time investment.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system performs preliminary analysis of search query co-occurrences to pre-identify potential inappropriate queries before formal classifier training. This preliminary action creates a ready-to-use training database, saving significant time during the actual classifier development phase.

Inventive Principle:

Principle #10Preliminary action

2Productivity

If co-occurrence analysis is used to identify candidate queries, then the database expansion efficiency is improved, but the risk of false positives increases

Engineering Contradiction:

Improvedatabase expansion efficiencyVSAvoidquery classification accuracy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system implements feedback loops where the classifier continuously learns from identified candidate queries, adjusting its parameters to reduce false positives. The co-occurrence analysis results are fed back into the training process, refining the classifier's ability to distinguish true inappropriate queries from false positives.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system intentionally casts a wide net with co-occurrence analysis, identifying more candidate queries than may ultimately be confirmed as inappropriate. This excessive action ensures high recall, and the classifier then filters these candidates to achieve the desired precision, balancing efficiency and reliability.

Inventive Principle:

Principle #16Partial or excessive action

3Object-affected harmful factors

If the classifier is trained on a comprehensive database of inappropriate content queries, then the content filtering effectiveness is improved, but the complexity of the system increases

Engineering Contradiction:

Improveinappropriate content exposureVSAvoidsystem complexity

Core Design Contradiction:

Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The system segments the inappropriate content detection task into multiple components: co-occurrence analysis for candidate identification, classifier training for pattern recognition, and iterative refinement for optimization. This segmentation makes the complex system more manageable and maintainable while preserving comprehensive filtering capability.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The co-occurrence analysis acts as an intermediary between raw search queries and the final classifier decision. It pre-processes and filters queries before they reach the classifier, reducing the complexity of the classification task while maintaining comprehensive content filtering effectiveness.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS9959354B2Utilizing user co-search behavior to identify search queries seeking inappropriate content

Publication Date: 2018.05.01 GOOGLE LLC

AI summary

A system and method for training a search query classifier may be used to develop a large database of search queries used to access inappropriate sensitive or offensive content. A database of well-known and frequently-used search queries for accessing inappropriate sensitive content is expanded by monitoring additional search queries received from a user within a preset time period of the user submitting one of the well-known and frequently-used search queries. The additional search queries received from a user are further evaluated to determine if they are likely associated with inappropriate sensitive or offensive content.