GNEM: A Generic One-to-Set Neural Entity Matching Framework

Chen, Runjin; Shen, Yanyan; Zhang, Dongxiang

doi:10.1145/3442381.3450119

Cited by 8 publications

(3 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The resulting index offers very good querying times with the tradeoff of a large overhead in building the index. In our experiments, we used the implementation provided by FAISS 5 . First, we vectorize and index all input entities.…”

Section: Evaluated Tasks and Methodologymentioning

confidence: 99%

“…The first such work is EMTransformer [3], which considered BERT and its three main variants: XLNet, RoBERTa and DistilBERT. The same models are used by GNEM [5], which extends EMTransformer through a graph that captures the relations between all candidate pairs that are given as input to matching. GNEM also applies this idea to DeepMatcher, in combination with FastText embeddings.…”

Section: Related Workmentioning

confidence: 99%

“…Over the years, numerous language models have been used in both ER steps. Blocking with GloVe [13] and FastText [56,67] and matching with GloVe [13,33], FastText [14,23,33,35,61,64,66] as well as BERT [46] and its main variants, i.e., XLNet, DistilBERT, RoBERTa and AlBERT [3,5,25,38]. However, there is no systematic analysis of their relative performance in these ER tasks.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Pre-Trained Embeddings for Entity Resolution: An Experimental Analysis

et al. 2023

View full text Add to dashboard Cite

Many recent works on Entity Resolution (ER) leverage Deep Learning techniques involving language models to improve effectiveness. This is applied to both main steps of ER, i.e., blocking and matching. Several pre-trained embeddings have been tested, with the most popular ones being fastText and variants of the BERT model. However, there is no detailed analysis of their pros and cons. To cover this gap, we perform a thorough experimental analysis of 12 popular language models over 17 established benchmark datasets. First, we assess their vectorization overhead for converting all input entities into dense embeddings vectors. Second, we investigate their blocking performance, performing a detailed scalability analysis, and comparing them with the state-of-the-art deep learning-based blocking method. Third, we conclude with their relative performance for both supervised and unsupervised matching. Our experimental results provide novel insights into the strengths and weaknesses of the main language models, facilitating researchers and practitioners to select the most suitable ones in practice.

show abstract

Section: Evaluated Tasks and Methodologymentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%