Search Result Clustering for Redundancy Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional online search methods primarily rely on relevance sorting, leading to redundancy and inefficiency as users face difficulty in finding diverse and relevant results, especially with ambiguous search queries, resulting in poor user experience and reduced search efficiency.
Innovation Solution
An online search method that determines an initial search result set and selects a final set based on both relevance and difference, using a semantic matrix to describe document attributes, clustering documents, and selecting representatives from each cluster to provide diverse and relevant results, incorporating spectral graph theory to construct similarity graphs and reduce redundancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If search results are sorted only by relevance to the query request, then the most relevant results are placed at the top, but redundant and similar results reduce user search efficiency and experience
Solution Approach 1:
The patent segments the search result set into multiple clusters based on semantic similarity. Each cluster groups together results with similar meanings, and the system selects representative results from each cluster. This segmentation approach maintains relevance accuracy while eliminating redundant results within the same semantic group, thereby improving search efficiency.
Solution Approach 2:
The patent introduces a new parameter 'degree of difference' in addition to the traditional relevance parameter. By changing the sorting criteria from solely relevance-based to a composite of relevance and difference, the system can distinguish between redundant similar results and diverse relevant results, resolving the contradiction between relevance accuracy and search efficiency.
2Reliability
If multiple highly relevant and similar results are returned, then relevance is maximized, but users find it difficult to maintain interest and may ignore other candidate results
Solution Approach 1:
The patent applies local quality by ensuring that within each semantic cluster, the selected representative result has high relevance quality, while the collection of representatives across different clusters provides diverse quality. This local optimization within clusters and global diversification across clusters maintains reliability while improving user engagement through variety.
Solution Approach 2:
Instead of selecting results solely based on highest relevance (traditional approach), the patent inverts the selection criterion by choosing results with the greatest 'degree of difference' from already selected results. This inversion ensures diversity while maintaining relevance, thereby improving user engagement without sacrificing reliability.
3Ease of operation
If keyword search is used with ambiguous phrases, then operation simplicity is maintained, but accurate understanding of user search intention becomes difficult
Solution Approach 1:
The patent introduces semantic clustering as an intermediary process between the user's simple keyword query and the final result selection. The clustering mechanism acts as a mediator that interprets ambiguous keywords by grouping results semantically, thereby improving search intention understanding while maintaining operation simplicity. The user still inputs simple keywords, but the system mediates the interpretation through semantic analysis.
Data Source
AI summary
The present invention relates to an online search method and device that accounts for relevance and difference, wherein the method comprises: determining an initial search result set according to a query request input by a user, and selecting a plurality of search results with the greatest degrees of difference from the initial search result set, as a final search result set. The specific process comprises: initially obtaining a plurality of candidate results according to the query request submitted by the user; using a semantic matrix to describe the attribute characteristics of documents corresponding to the candidate results; calculating similarities among the documents according to the attribute characteristics of the documents, and then clustering the candidate results into a plurality of cluster blocks; selecting a representative from each cluster block; and constructing the final search result set according to the selected representatives.