Query Suggestion Clustering via Proximity Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing online application stores face challenges in providing effective search suggestions for category-based application queries, as they often rely on general search engines that may not account for platform compatibility and application category specificity, leading to irrelevant results.
Innovation Solution
The implementation of a clustering technique that generates a proximity graph for category application queries, identifying clusters based on shared responsive applications and generating suggested queries based on similarity scores, ensuring that suggestions match the intended application categories and platform compatibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If general search engines are used for category-based application queries, then broad search coverage is achieved, but relevance to application categories and platform compatibility deteriorates
Solution Approach 1:
The patent segments the search space by creating clusters of category application queries based on shared responsive applications. Each cluster represents a specific category or subcategory, allowing the system to provide targeted suggestions for each segment rather than treating all queries uniformly. This segmentation enables both broad coverage across multiple categories and precise relevance within each category.
Solution Approach 2:
The system changes the parameter of query classification by stamping queries with category identifiers and organizing them into hierarchical clusters. By transforming queries from unstructured text into categorized, clustered data structures with similarity scores, the system maintains versatility across diverse query types while improving relevance through category-specific suggestion generation.
2Measurement precision
If category-specific clustering and proximity graphs are implemented, then query suggestion relevance is improved, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-processing category application queries to stamp them with category identifiers and pre-cluster them into groups based on shared responsive applications. The proximity graphs and similarity scores are computed in advance, allowing the system to quickly retrieve relevant suggestions without performing complex calculations in real-time, thus managing complexity while maintaining high relevance.
3Measurement precision
If proximity graphs with similarity scores are generated for all query pairs, then suggestion accuracy is improved, but computational resources increase
Solution Approach 1:
The system applies local quality by generating proximity graphs and similarity scores only within local clusters of related queries rather than computing all possible query pairs globally. Each cluster's proximity graph focuses on local relationships among queries with shared responsive applications, reducing overall computational energy while maintaining accuracy for category-specific suggestions.
Data Source
AI summary
A plurality of category application queries can be processed, including identifying an associated list of results for each of the category application queries, with each list of results listing one or more applications. A cluster of the category application queries that each have a category in common in an application category stamp can be identified. A proximity graph for the category application queries in the cluster can be generated. The proximity graph can indicate similarity scores based at least in part on numbers of applications in common for lists of results associated with the category application queries in pairs of queries in the cluster. Lists each having one or more suggested alternative category application queries for category application queries in the cluster can be generated, with the generating of the lists of one or more suggested alternative category application queries being based at least in part on the similarity scores.