Distributed representations of tuples for entity resolution

Abstract

Despite the efforts in 70+ years in all aspects of entity resolution (ER), there is still a high demand for democratizing ER - by reducing the heavy human involvement in labeling data, performing feature engineering, tuning parameters, and defining blocking functions. With the recent advances in deep learning, in particular distributed representations of words ( a.k.a . word embeddings), we present a novel ER system, called D eep ER, that achieves good accuracy, high efficiency, as well as ease-of-use ( i.e ., much less human efforts). We use sophisticated composition methods, namely uni- and bi-directional recurrent neural networks (RNNs) with long short term memory (LSTM) hidden units, to convert each tuple to a distributed representation ( i.e ., a vector), which can in turn be used to effectively capture similarities between tuples. We consider both the case where pre-trained word embeddings are available as well the case where they are not; we present ways to learn and tune the distributed representations that are customized for a specific ER task under different scenarios. We propose a locality sensitive hashing (LSH) based blocking approach that takes all attributes of a tuple into consideration and produces much smaller blocks, compared with traditional methods that consider only a few attributes. We evaluate our algorithms on multiple datasets (including benchmarks, biomedical data, as well as multi-lingual data) and the extensive experimental results show that D eep ER outperforms existing solutions.

Dual-objective fine-tuning of BERT for entity matching

“…Entity matching methods can broadly be divided into rule-based, crowd-based, and machine learning-based methods [5,6,14]. Since 2018, an increasing number of neural network-based matching methods [13,23,30] have been proposed and have pushed the state-of-the-art performance especially for textual entity matching tasks [1]. We include Deepmatcher [23] into our experiments as an example of one of the initial neural network based matching systems.…”

Deep learning based approach to unstructured record linkage

“…Distributed representation of records (DeepER). This is a recently proposed approach which applies a distributed representation of words (Ebraheem et al , 2018) for constructing a distributed representation of records. For each token (word) within an attribute value its distributed representation is obtained from one of the pre-trained embedding dictionaries.…”

“…In particular, the application of machine learning (ML) offers a promising approach, which can be applied as an alternative to manual rule building (Köpcke et al , 2010). However, the existing ML-based approaches to RL are based on the assumption that the data obtained from different sources is structured and represented by overlapping sets of attributes (Ebraheem et al , 2018; Elfeky et al , 2002; Jurek et al , 2017; Kejriwal and Miranker, 2015; Ngomo and Lyko, 2013; Schneider et al , 2018; Sherif et al , 2017; Wang et al , 2015). This is very restrictive in terms of real world applications, given the increasing number of unstructured data sources such as social media channels, for example.…”

Deep Learning for Semantic Matching: A Survey

“…Then the matching step determines if each pair in the candidate set is a match. To our knowledge, as of November 2018 there have been only two published work on entity matching using deep learning: DeepER [33] and DeepMatcher [72]. We now describe both.…”

“…Finally, given a set of tuples (e.g., the union of the two tables to be matched), we pass each tuple through all L hash tables to obtain a list of blocks. Then the candidate set for matching consists of all tuple pairs that appear together in at least one block (there are pruning strategies to further reduce the candidate set size, see [33]).…”