Rule-Based Data Ranking with Prototype Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for ranking large datasets, such as fraud detection in credit card transactions, are inefficient and error-prone, particularly in unbalanced data distributions, as they lack human-understandable automated ranking solutions.
Innovation Solution
A computer-implemented method for ranking data using learned rules, which generates a rule set from a training dataset, creates prototypes for each rule, and scores target data records based on proximity to these prototypes, focusing on the F-measure to prioritize the relevant class, thereby providing a more granular and precise ranking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated ranking methods such as neural networks and support vector machines are used, then ranking efficiency is improved, but human understandability deteriorates
Solution Approach 1:
The patent replaces complex automated ranking methods (neural networks, support vector machines) with a rule-based system that uses inductively learned rules. This substitution maintains automated processing efficiency while producing human-understandable rules that can be interpreted and explained, thus resolving the contradiction between productivity and understandability.
2Loss of information
If manual ranking of large data amounts is performed, then human understandability is maintained, but productivity deteriorates and error rate increases
Solution Approach 1:
The patent implements a self-service system where the computer automatically generates and applies ranking rules without human intervention. The inductive rule learning algorithm autonomously processes the training data, generates ranking rules, and applies them to target data, maintaining both high productivity and human understandability of the generated rules.
3Loss of information
If rule induction algorithms are used for classification, then human understandability is improved, but their application to ranking deteriorates
Solution Approach 1:
The patent extends rule induction algorithms from their traditional classification function to also perform ranking functions. By modifying the rule induction process to generate rules with ranking capabilities and introducing a ranking function that evaluates data items based on these rules, the system achieves multi-functionality, making rule induction applicable to both classification and ranking tasks.
4Productivity
If automated ranking methods are used, then productivity is improved, but measurement precision deteriorates in unbalanced data distributions
Solution Approach 1:
The patent applies local quality by optimizing the rule induction process specifically for unbalanced data distributions. The system adjusts the rule generation and evaluation metrics to account for class imbalance, ensuring that ranking precision is maintained even when the data distribution is skewed. This localized optimization for unbalanced data resolves the contradiction between productivity and measurement precision in such scenarios.
Data Source
AI summary
Systems, methods and computer program products for the ranking of a target data set based on learned rules are disclosed. One embodiment is a method that includes generating a learned rule set from a training data record set, creating at least one prototype for each rule in the learned rule set to generate a prototype set, and ranking the target data record set using learned rule set and the prototype set. The generating of a learned rule set includes dividing the training data record set to a positive class and a negative class, and deriving the learned rule set for the positive class. Learning of rules includes deriving the most general projected rules with respect to remaining training data and then refining those rules, eventually selecting the best rules using an F-measure.