[Paper Review] Unsupervised String Transformation Learning for Entity Consolidation

Dong Deng, Wenbo Tao+8 more|arXiv (Cornell University)|Sep 29, 2017

Data Quality and Management3 citations

TL;DR

This paper proposes an unsupervised string transformation learning method for entity consolidation that groups variant attribute values (e.g., 'Mary Lee' and 'Lee, Mary') into transformation-equivalent pairs using clustering, then validates these groups via minimal human feedback. In a real-world dataset of 17,497 records, the method achieved 75% recall and 99.5% precision with only 100 human-verification questions, significantly outperforming state-of-the-art data wrangling tools.

ABSTRACT

Data integration has been a long-standing challenge in data management with many applications. A key step in data integration is entity consolidation. It takes a collection of clusters of duplicate records as input and produces a single "golden record" for each cluster, which contains the canonical value for each attribute. Truth discovery and data fusion methods, as well as Master Data Management (MDM) systems, can be used for entity consolidation. However, to achieve better results, the variant values (i.e., values that are logically the same with different formats) in the clusters need to be consolidated before applying these methods. For this purpose, we propose a data-driven method to standardize the variant values based on two observations: (1) the variant values usually can be transformed to the same representation (e.g., "Mary Lee" and "Lee, Mary") and (2) the same transformation often appears repeatedly across different clusters (e.g., transpose the first and last name). Our approach first uses an unsupervised method to generate groups of value pairs that can be transformed in the same way (i.e., they share a transformation). Then the groups are presented to a human for verification and the approved ones are used to standardize the data. In a real-world dataset with 17,497 records, our method achieved 75% recall and 99.5% precision in standardizing variant values by asking a human 100 yes/no questions, which completely outperformed a state of the art data wrangling tool.

Motivation & Objective

• To address the challenge of standardizing variant attribute values in duplicate record clusters before entity consolidation.
• To reduce human effort in verifying string transformations by grouping semantically equivalent value pairs using unsupervised learning.
• To improve the accuracy of downstream entity consolidation methods by standardizing variant formats prior to golden record generation.
• To develop a scalable, human-in-the-loop approach that minimizes manual effort while maximizing transformation correctness.

Proposed method

• The method first extracts all value pairs from the same cluster as candidate replacements.
• It applies an unsupervised clustering algorithm to group candidate replacements that share the same transformation pattern (e.g., name reordering, abbreviation expansion).
• Groups are ranked by size and presented to a human for verification using a yes/no interface.
• Only groups marked as correct by the human are used to apply standardized transformations to the original data.
• The approach is robust to small numbers of incorrect groups, allowing high precision even with limited human feedback.
• The final standardized data enables more accurate entity consolidation using truth discovery, data fusion, or MDM systems.

Experimental results

Research questions

• RQ1 Can unsupervised clustering of string transformation candidates effectively group variant values that share the same transformation logic?
• RQ2 How can human feedback be minimized while maintaining high precision in identifying valid transformation groups?
• RQ3 To what extent does standardizing variant values improve the performance of downstream entity consolidation methods?
• RQ4 How does the proposed method compare to state-of-the-art data wrangling tools in terms of accuracy and human effort?

Key findings

• The method achieved 75% recall and 99.5% precision in standardizing variant values using only 100 human-verification questions on a real-world dataset of 17,497 records.
• The approach significantly outperformed a state-of-the-art data wrangling tool in both accuracy and efficiency.
• The unsupervised clustering of transformation-equivalent value pairs effectively captured common patterns such as name reordering and abbreviation expansion.
• The method demonstrated robustness to minor errors in group verification, maintaining high precision even with limited human input.
• By standardizing variant values before entity consolidation, the method enables better performance of truth discovery and MDM systems.
• The human-in-the-loop design minimized labor while achieving high-quality standardization, proving effective for real-world data integration pipelines.