Towards Universal Dense Blocking for Entity Resolution

TL;DR

This work introduces UniBlocker, a universal dense blocker for entity resolution that is pre-trained on a large, domain-diverse tabular corpus using self-supervised contrastive learning. By adopting a novel record-embedding and data paraphrasing strategy, UniBlocker achieves transferable blocking performance across multiple domains without domain-specific fine-tuning, outperforming prior self-supervised dense blockers and rivaling sparse methods. A new universal blocking benchmark spanning 19 datasets across 9 domains and 4 scenarios demonstrates the method's transferability and scalability, while ablations validate the contributions of open-domain pre-training, record embeddings, and paraphrasing techniques. The results suggest that universal, domain-agnostic dense blocking can be competitive with, and complementary to, existing sparse blocking approaches, with strong practical implications for scalable ER pipelines.

Abstract

Blocking is a critical step in entity resolution, and the emergence of neural network-based representation models has led to the development of dense blocking as a promising approach for exploring deep semantics in blocking. However, previous advanced self-supervised dense blocking approaches require domain-specific training on the target domain, which limits the benefits and rapid adaptation of these methods. To address this issue, we propose UniBlocker, a dense blocker that is pre-trained on a domain-independent, easily-obtainable tabular corpus using self-supervised contrastive learning. By conducting domain-independent pre-training, UniBlocker can be adapted to various downstream blocking scenarios without requiring domain-specific fine-tuning. To evaluate the universality of our entity blocker, we also construct a new benchmark covering a wide range of blocking tasks from multiple domains and scenarios. Our experiments show that the proposed UniBlocker, without any domain-specific learning, significantly outperforms previous self- and unsupervised dense blocking methods and is comparable and complementary to the state-of-the-art sparse blocking methods.

Figures (6)

• Figure 1: Illustration of one-model-per-dataset paradigm and one-model-all-datasets paradigm for dense blocking.
• Figure 2: Overview of UniBlocker pre-training procedure.
• Figure 3: Record input representation. Record embeddings are the sum of token embeddings of attribute value, attribute name embeddings and relative position embeddings.
• Figure 4: Average blocking times for varying approaches on scalability datasets with different record sizes.
• Figure 5: Average number of tokens using different tokenizers with increasing entity record size on two scalability datasets.

Theorems & Definitions (1)

• Definition 1: Blocking