Search Result Ranking Using Matrix Framework and Confidence Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engines using the Vector Space Model face inefficiencies due to high computational complexity in matrix operations for determining document relevancy, especially with large document collections, leading to prolonged processing times.
Innovation Solution
A method involving the generation of matrices to represent terms and documents, sorting based on preferences, calculating overlap measures, and computing cumulative confidence scores to rank search results efficiently, reducing computational complexity through optimized matrix operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional matrix multiplication operations are used to compute relevancy scores in the Vector Space Model, then document relevancy can be determined, but computational complexity increases to O(mn²) and processing time extends to minutes or hours
Solution Approach 1:
The patent segments the large TxD matrix into smaller sub-matrices or blocks, processing them in manageable chunks rather than performing monolithic matrix multiplication. This division reduces the computational burden from O(mn²) to approximately O(k·mn²) where k < 1, enabling faster processing while maintaining relevancy score accuracy through systematic aggregation of block results
Solution Approach 2:
The patent performs preliminary actions by pre-computing and storing term frequency vectors, document vectors, and similarity matrices before actual search operations. These pre-computed structures are cached and reused across multiple queries, eliminating redundant calculations and significantly reducing processing time for subsequent search operations
2Reliability
If the complete TxD matrix is processed for all documents and terms, then comprehensive search coverage is achieved, but computational time increases significantly for large collections
Solution Approach 1:
The patent extracts and processes only the relevant sub-sections of the TxD matrix corresponding to the query terms and candidate documents, rather than processing the entire matrix. This selective extraction focuses computational resources on the subset of data that actually contributes to the search results, maintaining completeness for relevant items while eliminating waste on irrelevant portions
Solution Approach 2:
The patent implements partial action by computing relevancy scores for only the top-k candidate documents that pass initial filtering thresholds, rather than scoring all documents in the collection. This approach achieves practical search completeness by identifying the most relevant results without the excessive computational cost of evaluating every document
Data Source
AI summary
A method and apparatus are provided for ranking documents according to relevancy scoring. In one implementation, a computer-implemented method is provided for receiving search results identifying a plurality of documents resulting from a search, the plurality of documents containing one or more words. The method generates a first matrix containing a term column and a document column, wherein at least one row of the first matrix correlates one of the plurality of documents with one of the terms. The method selects a sort preference, and sorts the two-column matrix according to the sort preference. The method further generates a second matrix containing values representing a measure of overlap between the plurality of documents and the terms. The method further calculates cumulative confidence scores according to the values of the second matrix and ranks the search results according to the cumulative confidence scores.