Entity matching (EM) is a fundamental task in data integration and analytics, essential for identifying records that refer to the same real-world entity across diverse sources. In practice, datasets often differ widely in structure, format, schema, and semantics, creating substantial challenges for EM. We refer to this setting as Heterogeneous EM (HEM) . This survey offers a unified perspective on HEM by introducing a taxonomy, grounded in prior work, that distinguishes two primary categories– representation and semantic heterogeneity –and their subtypes. The taxonomy provides a systematic lens for understanding how variations in data form and meaning shape the complexity of matching tasks. We then connect this framework to the FAIR principles – Findability , Accessibility , Interoperability , and Reusability –demonstrating how they both reveal the challenges of HEM and suggest strategies for mitigating them. Building on this foundation, we critically review recent EM methods, examining their ability to address different heterogeneity types, and conduct targeted experiments on state-of-the-art models to evaluate their robustness and adaptability under semantic heterogeneity. Our analysis uncovers persistent limitations in current approaches and points to promising directions for future research, including multimodal matching, human-in-the-loop workflows, deeper integration with large language models and knowledge graphs, and fairness-aware evaluation in heterogeneous settings.

Entity matching (EM), also known as record linkage, is crucial in data integration, cleaning, and knowledge base construction. Modern matching techniques leverage deep learning and pre-trained language models (PLMs) to effectively identify matching records, showcasing significant advancements over traditional methods. However, certain critical matching aspects have received limited attention in these techniques. They heavily rely on PLMs’ encodings and face challenges in integrating external sources of knowledge to enhance matching accuracy. Additionally, these techniques often lack transparency, impeding users’ understanding of the underlying rationale for matching decisions. Furthermore, they exhibit limitations and decreased performance in handling heterogeneous records from datasets with diverse schemas. This paper presents EXKG, a novel technique that addresses these challenges and effectively matches heterogeneous records with varying attributes. EXKG combines the power of knowledge graphs (KGs) and PLMs to perform record linkage while offering explanatory insights into the matching results. We demonstrate that EXKG achieves competitive performance through experimental studies compared to state-of-the-art matching techniques. As a by-product, our solution generates explanations that give end users a comprehensive understanding of the matching process. We evaluate the quality of these explanations by using a user study and show they empower end users to make informed decisions

Computing robust and accurate quality scores for users and items in online review systems is critical, since scores directly reflect the community-wide belief about their quality. A broad range of methods have been proposed to compute rating scores, including simple aggregation, weighted aggregation, and iterative techniques, where the latter provides relatively accurate results. However, there are still serious challenges to address, especially in terms of time complexity, accuracy, and robustness against manipulation. In this article, we propose an adaptive quality assessment framework that computes dependable and accurate quality scores for users and items. The proposed method is a semi-iterative weighted aggregation technique in which, a novel approach is used to assign weights to received reviews. The weight depends on two parameters: similarity of reviews, and review prediction. In review prediction, we utilize a combination of online machine learning and collaborative filtering to predict the review expected from the user. The intuition behind using online learning is its ability to obtain lower time complexity in comparison with batch learning. We evaluate our proposed model using a real-word dataset, and compare it with two related approaches. Results show the superiority of our proposed approach, in terms of accuracy and robustness against manipulation.