Beyond Scale and Generation: Understanding Language Model-based Entity Matching
2026-07-27 • Databases
DatabasesComputation and LanguageMachine Learning
AI summaryⓘ
The authors studied how well three types of language model setups (bi-encoder, cross-encoder, and generative matcher) work for identifying if records refer to the same real-world thing. They did many experiments controlling for model type, size, and variant to see what really helps performance. They found that for bi-encoders, the model variant matters a lot, while cross-encoders usually perform better because they look at record pairs together. Generative matchers excel mostly when dealing with data changes or new types of records. The authors also note that simply using bigger models doesn’t always improve results because bigger models may rely on shortcuts.
entity matchingbi-encodercross-encodergenerative matcherlanguage modelsmodel variantdistribution shiftcross-dataset transferrepresentation geometryshortcut learning
Authors
Zeyu Zhang, Xue Li, Iacer Calixto, Paul Groth, Sebastian Schelter
Abstract
Entity matching identifies records that refer to the same real-world entity. Language models can be adapted to this task through bi-encoder, cross-encoder, and generative matcher architectures. However, prior studies often conflate matcher architecture with differences in model backbone, model variant(reflecting different pretraining objectives), and model size, making it difficult to isolate the sources of performance gains. We address this issue through a controlled factorial study spanning three matcher architectures, three model variants and three model sizes from the Qwen3 family, and nine datasets, totaling 1,215 fine-tuning runs. We also evaluate cross-dataset transferability and computational cost. Our results show that model variant is critical for bi-encoders: embedding-oriented variants provide stronger initialization and more favorable representation geometry predictive of downstream matching performance. Cross-encoders retain a consistent advantage over bi-encoders because they jointly encode record pairs rather than representing each record independently, although larger models partially narrow this gap. Generative matchers do not universally outperform cross-encoders. Instead, their advantages concentrate under distribution shift, including subtle unseen differences in record schemas and cross-dataset transfer. We further find that larger models rely more heavily on shortcut learning and therefore do not necessarily perform better. These findings clarify the factors underlying performance differences across matcher architectures and motivate future research and benchmark designs that better disentangle architectural choices from model-level factors while explicitly evaluating distribution shift and cross-dataset transferability. We release our experimental results, code, training scripts, and evaluation data at https://github.com/Jantory/llm-trained-matcher.