Heterogeneous Committee-Based Active Learning for Entity Resolution (HeALER)

Chen, Xiao; Xu, Yuanyuan; Broneske, David; Durand, Gabriel Campero; Zoun, Roman; Saake, Gunter

doi:10.1007/978-3-030-28730-6_5

Cited by 8 publications

(9 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Meduri et al [32] provide an in-depth comparison of various matchers and example selectors but neither consider TPLMs nor address how to learn a blocker. While HEALER [10] attempts to improve upon Mozafari et al [33]'s committee-based approach to ER by including different kinds of matchers, it does not consider neural networks or TPLMs in its heterogeneous committee. Alongside AL for ER, another line of work attempts to solve ER by crowd-sourcing [53].…”

Section: Active Learning For Entity Resolutionmentioning

confidence: 99%

“…Popular choices for matcher includes support vector machines [45], random forests [32], and neural networks [23]. Popular choices for selector includes query-by-committee [16,49] which has seen wide usage in ER [10,33] and uncertainty sampling [23].…”

Section: Introductionmentioning

confidence: 99%

“…This design affords sufficient flexibility so that we can utilize separate loss functions to train either matcher or the later encoders (e.g., by freezing the matcher's parameters). As mentioned before, while previous works on AL and ER have attempted to learn committees instead of a single classifier [10,32,33,45], none of them consider TPLMs since these were only recently introduced. Even in the broader research community, learning a committee on top of TPLMs is an open problem, to the best of our knowledge.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Deep Indexed Active Learning for Matching Heterogeneous Entity Representations

Jain,

Sarawagi,

Sen

2021

Preprint

View full text Add to dashboard Cite

Given two large lists of records, the task in entity resolution (ER) is to find the pairs from the Cartesian product of the lists that correspond to the same real world entity. Typically, passive learning methods on tasks like ER require large amounts of labeled data to yield useful models. Active Learning is a promising approach for ER in low resource settings. However, the search space, to find informative samples for the user to label, grows quadratically for instance-pair tasks making active learning hard to scale. Previous works, in this setting, rely on hand-crafted predicates, pre-trained language model embeddings, or rule learning to prune away unlikely pairs from the Cartesian product. This blocking step can miss out on important regions in the product space leading to low recall. We propose DIAL, a scalable active learning approach that jointly learns embeddings to maximize recall for blocking and accuracy for matching blocked pairs. DIAL uses an Index-By-Committee framework, where each committee member learns representations based on powerful transformer models. We highlight surprising differences between the matcher and the blocker in the creation of the training data and the objective used to train their parameters. Experiments on five benchmark datasets and a multilingual record matching dataset show the effectiveness of our approach in terms of precision, recall and running time. 1

show abstract

Section: Active Learning For Entity Resolutionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep Indexed Active Learning for Matching Heterogeneous Entity Representations

Jain,

Sarawagi,

Sen

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In each active learning iteration, one or more informative record pairs are selected and provided to a human annotator for labeling. One way to measure the degree of informativeness is to calculate the disagreement among the predictions of a classifier ensemble known as the Query-by-Committee strategy [2,19,21]. The most informative record pairs are the ones that cause the highest disagreement among the members of the committee.…”

Section: Introductionmentioning

confidence: 99%

“…To circumvent the cold start problem, existing active learning methods for entity resolution require the human annotator to label a small set of record pairs. This set is either selected randomly [9,18] or based on the distribution of pre-calculated similarity scores [2,19] before the active learning starts. However, solving the cold start problem by manually annotating a subset of the data is contradicting the main principle of active learning, that of minimizing the human labeling effort.…”

Section: Introductionmentioning

confidence: 99%

Unsupervised Bootstrapping of Active Learning for Entity Resolution

2020

View full text Add to dashboard Cite

Entity resolution is one of the central challenges when integrating data from large numbers of data sources. Active learning for entity resolution aims to learn high-quality matching models while minimizing the human labeling effort by selecting only the most informative record pairs for labeling. Most active learning methods proposed so far, start with an empty set of labeled record pairs and iteratively improve the prediction quality of a classification model by asking for new labels. The absence of adequate labeled data in the early active learning iterations leads to unstable models of low quality which is known as the cold start problem. In our work we solve the cold start problem using an unsupervised matching method to bootstrap active learning. We implement a thresholding heuristic that considers pre-calculated similarity scores and assigns matching labels with some degree of noise at no manual labeling cost. The noisy labels are used for initializing the active learning process and throughout the whole active learning cycle for model learning and query selection. We evaluate our pipeline with six datasets from three different entity resolution settings using active learning with a committee-based query strategy and show it successfully deals with the cold start problem. Comparing our method against two active learning baselines without bootstrapping, we show that it can additionally lead to overall improved learned models in terms of F1 score and stability.

show abstract

Graph-Boosted Active Learning for Multi-source Entity Resolution

Primpeli

Bizer

2021

The Semantic Web – ISWC 2021

View full text Add to dashboard Cite

Heterogeneous Committee-Based Active Learning for Entity Resolution (HeALER)

Cited by 8 publications

References 19 publications

Deep Indexed Active Learning for Matching Heterogeneous Entity Representations

Deep Indexed Active Learning for Matching Heterogeneous Entity Representations

Unsupervised Bootstrapping of Active Learning for Entity Resolution

Graph-Boosted Active Learning for Multi-source Entity Resolution

Contact Info

Product

Resources

About