Multi-Engine Search Result Ranking with Fully Connected Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing election algorithms for combining results from multiple search engines, such as the Schulze method, scale poorly with the number of candidates, requiring cubic time complexity and often necessitating multiple compute cores, which is inefficient and costly.
Innovation Solution
A method using a fully-connected neural network to process ballots from multiple search engines, generating normalized matrices and vectors to rank candidates efficiently, achieving linear or quadratic complexity and enabling single-core processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If election algorithms like the Schulze method are used to combine results from multiple search engines, then the ranking accuracy is improved, but the computational time increases cubically with the number of candidates
Solution Approach 1:
The patent replaces the mechanical election algorithm (Schulze method) with a neural network-based system. The neural network is trained to predict candidate rankings by processing search engine results, substituting the cubic-time complexity mechanical voting system with a more efficient machine learning approach that scales better while maintaining ranking accuracy.
Solution Approach 2:
The patent applies preliminary action by pre-training the neural network on historical search engine results and candidate data before actual use. This training phase allows the system to learn optimal ranking patterns in advance, so that during actual operation, rankings can be generated quickly without performing complex real-time calculations.
2Productivity
If multiple compute cores are used to run election algorithms, then the processing speed is improved, but the hardware cost and system complexity increase
Solution Approach 1:
The patent substitutes the parallel-processing requirement of traditional election algorithms with a neural network architecture that can process search engine results more efficiently. The neural network's structure allows it to handle the ranking computation with reduced hardware demands, potentially running on single-core processors while maintaining or improving processing speed.
Solution Approach 2:
The patent changes the computational parameters by transitioning from exact election algorithm calculations to probabilistic neural network predictions. This parameter change allows the system to accept approximate results with high confidence, reducing the need for multiple compute cores and complex hardware configurations.
3Reliability
If the number of search engines is increased to improve result completeness, then the coverage is improved, but the computational complexity of combining results increases
Solution Approach 1:
The patent applies universality by designing a neural network that can process results from any number of search engines through a unified architecture. The same neural network model handles inputs from two search engines as well as from many search engines, eliminating the need for different algorithms or increased complexity when scaling the number of search engines.
Solution Approach 2:
The patent changes the approach from combinatorial algorithm complexity to fixed neural network complexity. Regardless of how many search engines provide input, the neural network processes them through the same layers with the same number of parameters, preventing algorithmic complexity from increasing with the number of search engines.
Data Source
AI summary
In an embodiment, a plurality of ballots (1) for a plurality of candidates, (2) generated by a plurality of search engines, and (3) for a job description are received. Each ballot from the plurality of ballots is generated by a search engine from the plurality of search engines different than remaining search engines from the plurality of search engines. A mathematical representation that indicates, for each candidate from the plurality of candidates, how many other candidates from the plurality of candidates that candidate was ranked higher than in the plurality of ballots is generated. A final ballot ranking the plurality of candidates is generated using a trained statistical model and based on the mathematical representation. A candidate from the plurality of candidates for the job description is identified based on the final ballot.