Active Learning Record Matching with Precision Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing active learning record matching techniques lack control over classifier quality, are unpredictable, and do not scale well to large inputs, as they require manual selection of labeled examples and iterate over all record pairs, leading to inefficiencies in precision and recall.
Innovation Solution
An active learning record matching system that allows users to specify a precision threshold, uses a blocking technique to limit the search space, and employs a learning method to generate a record matching package with precision greater than or equal to the threshold, while estimating precision and recall using a binary blocking function and randomly selecting labeled examples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing active learning record matching techniques are used, then the system can learn from labeled examples, but the quality of the learned classifier cannot be controlled and precision may decrease when more labeled examples are provided
Solution Approach 1:
The system performs preliminary actions by computing an upper bound on precision before full learning occurs. The precision upper bound computation analyzes the candidate record pairs and blocking function in advance to establish a theoretical maximum precision level, allowing the system to predict whether the learned classifier will meet quality requirements before actual labeling and learning take place
Solution Approach 2:
The system implements feedback by using the computed precision upper bound to guide the active learning process. If the upper bound indicates that precision requirements cannot be met, the system can adjust its approach by selecting different blocking functions or requiring additional labeled examples, creating a feedback loop that ensures quality control
2Measurement precision
If existing active learning record matching techniques iterate over all record pairs to select labeled examples, then they can comprehensively evaluate candidates, but they do not scale to large inputs due to quadratic complexity
Solution Approach 1:
The system applies segmentation by dividing the record matching task into two stages: first using a blocking function to partition records into candidate pairs, then applying the learned classifier only to these candidates. This segmentation reduces the problem size from all possible record pairs to a manageable subset, enabling scaling to large inputs while maintaining evaluation quality
Solution Approach 2:
The system uses partial action by computing the precision upper bound on a sampled subset of candidate pairs rather than exhaustively evaluating all pairs. This partial evaluation is sufficient to determine whether precision requirements can be met, avoiding the quadratic complexity of full evaluation while maintaining the ability to assess quality
Data Source
AI summary
An active learning record matching system and method for producing a record matching package that is used to identify pairs of duplicate records. Embodiments of the system and method allow a precision threshold to be specified and then generate a learned record matching package having precision greater than this threshold and a recall close to the best possible recall. Embodiments of the system and method use a blocking technique to restrict the space of record matching packages considered and scale to large inputs. The learning method considers several record matching packages, estimates the precision and recall of the packages, and identifies the package with maximum recall having precision greater than equal to the given precision threshold. A human domain expert labels a sample of record pairs in the output of the package as matches or non-matches and this labeling is used to estimate the precision of the package.