Query Classification Using Refinement-Based Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current internet search engines lack an effective method to classify queries, leading to inefficient retrieval of relevant resources and inadequate refinement of search results, which hampers user satisfaction and advertising relevance.
Innovation Solution
A system that classifies queries by determining refinements, calculating similarity using meta-terms, and applying category labels based on manually classified samples to aggregate and propagate classifications across similar queries, utilizing word-vector cosine similarity and term frequency-inverse document frequency scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If query classification is implemented using manual categorization only, then classification accuracy is improved, but processing time and scalability deteriorate
Solution Approach 1:
The system performs preliminary manual classification on a sample subset of queries to establish accurate category labels and refinement relationships. These pre-computed classifications and similarity metrics are then reused to automatically classify similar queries, avoiding the need for manual classification of every query while maintaining accuracy.
Solution Approach 2:
The system copies classification labels and category structures from manually classified sample queries to similar unclassified queries based on similarity metrics. Instead of manually classifying each query individually, the system replicates classifications from representative samples to analogous queries, significantly reducing processing time while preserving accuracy.
2Measurement precision
If all queries are manually classified, then classification quality is improved, but resource consumption and cost increase
Solution Approach 1:
The system applies partial manual classification only to a representative sample of queries rather than all queries. This partial action is sufficient to establish the classification framework and similarity metrics needed to automatically classify the remaining queries, reducing resource consumption while maintaining overall classification quality.
Solution Approach 2:
The system enables unclassified queries to classify themselves automatically by comparing them against the manually classified sample queries using similarity metrics. Each query determines its own category based on its similarity to classified samples, eliminating the need for manual intervention in every case while preserving classification quality.
3Productivity
If query refinements are not considered, then processing speed is improved, but classification accuracy deteriorates
Solution Approach 1:
The system extracts refinement terms from query refinements and uses them as meta-terms to calculate similarity between queries. By separating and reusing these refinement terms across multiple query comparisons, the system improves classification accuracy without proportionally increasing processing complexity, maintaining processing speed while enhancing accuracy.
4Productivity
If similarity calculation uses only basic term matching, then processing efficiency is improved, but classification precision deteriorates
Solution Approach 1:
The system changes the parameter used for similarity calculation from basic term matching to cosine similarity based on word vectors and TF-IDF weights. This parameter transformation captures semantic relationships and term importance more effectively, improving classification precision while maintaining computational efficiency through vector space operations.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for classifying queries. One of the methods includes the actions of obtaining a collection of search queries; for each query in the collection of search queries, determining whether other queries in the collection of search queries are refinements of the query, wherein the refinements of each query include the query terms and one or more additional terms not found in the query; determining a similarity between query pairs from the collection, wherein the similarity between a query pair is determined using the additional query terms from the refinements of the respective search queries of the query pair; obtaining a classified sample group of search queries using a plurality of category labels; and classifying each other search query of the collection of search queries having one or more refinements.