Entity Matching via Cross-Feature Computation and Feature Importance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for matching entities from distinct datasets require high computational resources, significant memory usage, and energy consumption, leading to increased latency and inefficiency, especially in real-time decision-making applications.
Innovation Solution
A computer-implemented method using cross-feature computation (CFC) and features importance computation (FIC) processes to compute cross-features and feature trend scores, allowing for efficient matching results without directly employing a machine learning model, leveraging Shapley values and Accumulated Local Effects (ALE) to estimate similarity scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used to compute similarity scores between entities, then matching accuracy is improved, but computational resources, memory usage, and energy consumption increase significantly
Solution Approach 1:
The patent segments the matching process into two distinct phases: an offline training phase where the machine learning model is trained on historical data to learn feature importance patterns, and an online inference phase where pre-computed feature importance weights are applied to new entity pairs without invoking the full ML model. This segmentation allows the system to achieve accurate matching results while minimizing real-time computational and energy costs.
Solution Approach 2:
The patent performs preliminary computation by pre-training the machine learning model offline and extracting feature importance weights (such as SHAP values or permutation importance) before actual matching operations. These pre-computed importance weights are stored and reused during online matching, eliminating the need to re-execute the computationally intensive ML model for each new entity pair, thus reducing energy consumption while maintaining matching accuracy.
2Measurement precision
If traditional machine learning models are used to compute similarity scores, then matching accuracy is improved, but latency increases due to processing vast amounts of data
Solution Approach 1:
The patent divides the matching system into an offline training component that performs heavy data processing to establish feature importance patterns, and an online matching component that applies these pre-established patterns efficiently. This segmentation transfers the time-consuming processing to the offline phase, enabling low-latency matching operations in real-time applications.
Solution Approach 2:
The system performs preliminary analysis by training the machine learning model and computing feature importance weights in advance on historical training data. During online operations, only lightweight computations using pre-computed weights are required, dramatically reducing latency for matching new entity pairs while preserving the accuracy benefits of the trained model.
3Measurement precision
If machine learning models are trained on large datasets to improve matching accuracy, then prediction quality is improved, but device complexity and computation needs increase
Solution Approach 1:
The patent extracts the essential knowledge from the trained machine learning model in the form of feature importance weights, which are then separated from the complex model structure. Only these extracted weights are retained and applied during online matching, eliminating the need to store and execute the entire complex ML model while preserving its predictive capabilities for the specific matching task.
Solution Approach 2:
Instead of using the full machine learning model during inference, the patent creates a simplified copy or representation of the model's knowledge through pre-computed feature importance weights. This lightweight copy captures the essential matching patterns learned during training, enabling accurate predictions without the computational overhead of the original complex model.
Data Source
AI summary
A computer-implemented method and system for matching entities from distinct datasets without directly employing a machine learning model. The method leverages cross-feature computation, features importance computation, and feature trend scores to efficiently and accurately predict matching scores between entities from a first dataset and a second distinct dataset. By reducing energy consumption, memory usage, and computation needs, the method enables faster decision-making, improved responsiveness, and streamlined matching processes across various domains, such as healthcare, finance, human resources management, or e-commerce applications. The method provides a scalable and adaptable solution that can be easily integrated into existing systems or workflows, promoting greater versatility and customization in the matching process and accommodating a wide range of applications and use cases.