Search Query Classification via Search Result Feature Vectors

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing search query classification methods face challenges in accurately determining user intent due to short and ambiguous queries, require extensive historical query logs that are time-consuming and costly to build, and struggle with updating classifications for new or old search terms.

Innovation Solution

A computer-implemented method generates feature vectors from search engine results for training queries, allowing for classification of search queries regardless of the availability of query logs, and dynamically updates classifications based on changing search engine results, using Uniform Resource Locators, titles, and content to determine the most relevant classifications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If historical query logs are used for search query classification, then classification accuracy can be improved, but the time and cost required to build and maintain these logs increases significantly

Engineering Contradiction:

Improveclassification accuracyVSAvoidtime to build query logs

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary classification by generating feature vectors from search engine results for training queries before actual classification is needed. This pre-computation of classification features eliminates the need to build extensive historical query logs, as the classification model is prepared in advance using readily available search results.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Instead of using actual historical query logs which require extensive collection and curation, the system creates synthetic training data by generating feature vectors from search engine results for selected training queries. This copying approach replicates the essential classification information without requiring real historical data accumulation.

Inventive Principle:

Principle #26Copying

2Measurement precision

If historical query logs are used for search query classification, then classification accuracy can be improved, but the cost of curating and maintaining these logs increases

Engineering Contradiction:

Improveclassification accuracyVSAvoidcost to build query logs

Core Design Contradiction:

Measurement precisionVSEase of manufacture

Solution Approach 1:

The system replaces expensive, hard-to-maintain historical query logs with inexpensive, easily generated feature vectors from current search engine results. These feature vectors can be quickly created and discarded or updated without the high curation and maintenance costs associated with building comprehensive query log databases.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The system uses its own search engine results to generate training data and feature vectors, eliminating the need to externally collect and curate historical query logs. The search engine serves itself by providing the data needed for classification model training and updating.

Inventive Principle:

Principle #25Self-service

3Stability of the object's composition

If traditional classification methods are used, then existing classifications can be maintained, but the system cannot quickly adapt to new classifications or changing user intent

Engineering Contradiction:

Improveclassification consistencyVSAvoidability to update classifications

Core Design Contradiction:

Stability of the object's compositionVSAdaptability or versatility

Solution Approach 1:

The classification system is made dynamic by continuously updating feature vectors based on current search engine results. As search results change over time, the feature vectors are regenerated, allowing the classification model to adapt to new classifications and changing user intent while maintaining consistency through the structured vector comparison approach.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system incorporates feedback loops where search engine results are continuously monitored and used to update feature vectors. This feedback mechanism allows the classification system to detect changes in search results and automatically adjust classifications, enabling rapid adaptation to new topics and changing user behavior patterns.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS8756218B1Query classification based on search engine results

Publication Date: 2014.06.17 GOOGLE LLC

AI summary

The technology described relates to performing query classification by using search engine results. In one aspect, a method includes identifying, from search results responsive to first search queries, first collections of features associated with classifications for the first search queries; in response to receiving a second search query, generating from search results responsive to the second search query a second collection of features associated with the second search query; and comparing the first collections of features to the second collection of features to determine a particular classification for the second search query.