A SURVEY ON APPROACHES FORSOLVING ENTITY MATCHINGPROBLEMS

The Hong Kong University of Science and Technology (Guangzhou)

Data Science and Analytics Thrust

PhD Qualifying Examination

By Mr. Yihao FU

Abstract

Entity matching (EM) is a fundamental task in data integration and information retrieval,
aimed at identifying and aligning entities across heterogeneous datasets. This survey categorizes the diverse methodologies employed in entity matching into four primary approaches:
rule-based methods, crowdsourcing techniques, traditional machine learning algorithms, and
deep learning frameworks.
The rule-based approach relies on predefined heuristics and expert knowledge to formulate
matching rules. While this method can achieve high precision in specific domains, it often
lacks scalability and adaptability when faced with new or evolving datasets. Consequently,
rule-based systems may struggle to maintain performance as the complexity of data increases.
In contrast, crowdsourcing leverages the collective intelligence of a diverse group of users
to validate and refine entity matches. This approach offers significant flexibility and can
adapt to various contexts, making it particularly useful in dynamic environments. However,
crowdsourced data may exhibit inconsistencies in quality due to varying levels of expertise
among contributors.
Traditional machine learning methods employ labeled training data to learn patterns for
entity matching. These techniques strike a balance between accuracy and generalizability,
allowing for the development of models that can handle a range of matching scenarios.
Nonetheless, they often require extensive feature engineering and may not fully capture the
complex relationships inherent in the data.
Deep learning techniques have recently emerged as powerful alternatives, utilizing neural
networks to automate feature extraction and model intricate entity relationships. These
methods have demonstrated superior performance on large-scale datasets and can effectively
handle unstructured data. However, deep learning approaches typically demand substantial computational resources and large volumes of training data, which may pose challenges for
smaller organizations or less resource-intensive applications.
This survey provides a comprehensive overview of these methodologies, highlighting their
strengths and limitations, and discusses future directions for research in entity matching.

PQE Committee

Chair of Committee: Prof. Nan TANG

Prime Supervisor: Prof. Wei WANG

Co-Supervisor: Prof. Wenjia WANG

Examiner: Prof. Zishuo DING