Dissimilar Data Correlation Using Symmetric Fit Score Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recruitment systems perpetuate bias, leading to sub-optimal filtering and ranking of candidates, especially from diverse communities, and fail to consider market dynamics, resulting in inefficient and unfair hiring processes.
Innovation Solution
A system using machine learning and deep learning with unbiased data to determine symmetric fit scores, incorporating integer optimization for fair and optimized hiring, addressing both asset and unit preferences while ensuring envy-free outcomes through a many-to-many stable correlation formulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional greedy recommendation systems are used to match candidates with jobs, then the strongest predicted fits are offered first, but this causes candidates from diverse communities to be ranked lower and marginalized individuals to be excluded
Solution Approach 1:
The patent inverts the traditional greedy matching approach by using many-to-many stable correlations that consider both candidate preferences and job preferences simultaneously. Instead of ranking candidates solely by predicted fit strength, the system computes stable matchings where no candidate-job pair would prefer each other over their current matches, thereby eliminating the bias against diverse candidates while maintaining high matching quality
Solution Approach 2:
The patent changes the fundamental parameter of matching from a single predicted fit score to a symmetric fit score that incorporates both candidate-job and job-candidate preferences. This parameter transformation enables the system to achieve both high matching quality and fairness by considering mutual preferences rather than unidirectional predictions
2Productivity
If automated filtering and pre-selection processes are used, then time and resources are saved, but bias is reproduced or exacerbated by algorithms
Solution Approach 1:
The patent introduces many-to-many stable correlations as an intermediary mechanism between automated filtering and final hiring decisions. This intermediary computes matchings that satisfy both efficiency requirements and fairness constraints by considering mutual preferences, thereby maintaining productivity while eliminating algorithmic bias in the filtering process
3Adaptability or versatility
If traditional recommendation systems match candidates with many job openings, then more opportunities are provided, but candidates apply to jobs for which they are not highly competitive, lengthening the recruitment process
Solution Approach 1:
The patent applies dynamic matching by computing many-to-many stable correlations that adapt to both candidate and job preferences. The system dynamically determines the optimal number of job recommendations for each candidate based on their specific preferences and competitiveness, rather than providing a fixed number of recommendations. This dynamic approach ensures candidates receive opportunities they can realistically compete for, reducing wasted applications and recruitment time
Data Source
AI summary
Systems and methods are disclosed for determining a plurality of best-fit correlations or matches between dissimilar data sets. An example method includes obtaining a first data set and a second data set from data sources. The method may include pre-processing the first data set to convert the received data into a standard format corresponding to attributes. A plurality of subsets corresponding to the second data set may be determined based on the attributes corresponding to the first data set. The method may include determining sets of fit scores individually corresponding to each of the one or more subsets of the first data set. The method may include determining the plurality of best-fit correlations or matches via an integer optimization model and based on the fit scores. The method may include displaying a list of the plurality of best-fit correlations or matches via a graphical user interface.