Recent studies find existing self-supervised speech encoders contain primarily acoustic rather than semantic information. As a result, pipelined supervised automatic speech recognition (ASR) to large language model (LLM) systems achieve state-of-the-art results on semantic spoken language tasks by utilizing rich semantic representations from the LLM. These systems come at the cost of labeled audio transcriptions, which is expensive and time-consuming to obtain. We propose a task-agnostic unsupervised way of incorporating semantic information from LLMs into self-supervised speech encoders without labeled audio transcriptions. By introducing semantics, we improve existing speech encoder spoken language understanding performance by over 10\% on intent classification, with modest gains in named entity resolution and slot filling, and spoken question answering FF1 score by over 2\%. Our unsupervised approach achieves similar performance as supervised methods trained on over 100 hours of labeled audio transcripts, demonstrating the feasibility of unsupervised semantic augmentations to existing speech encoders.
The paradigm of fine-tuning Pre-trained Language Models (PLMs) has been successful in Entity Matching (EM). Despite their remarkable performance, PLMs exhibit tendency to learn spurious correlations from training data. In this work, we aim at investigating whether PLM-based entity matching models can be trusted in real-world applications where data distribution is different from that of training. To this end, we design an evaluation benchmark to assess the robustness of EM models to facilitate their deployment in the real-world settings. Our assessments reveal that data imbalance in the training data is a key problem for robustness. We also find that data augmentation alone is not sufficient to make a model robust. As a remedy, we prescribe simple modifications that can improve the robustness of PLM-based EM models. Our experiments show that while yielding superior results for in-domain generalization, our proposed model significantly improves the model robustness, compared to state-of-the-art EM models.
Deep Learning revolutionizes almost all fields of computer science including data management. However, the demand for high-quality training data is slowing down deep neural nets' wider adoption. To this end, data augmentation (DA), which generates more labeled examples from existing ones, becomes a common technique. Meanwhile, the risk of creating noisy examples and the large space of hyper-parameters make DA less attractive in practice. We introduce Rotom, a multi-purpose data augmentation framework for a range of data management and mining tasks including entity matching, data cleaning, and text classification. Rotom features InvDA, a new DA operator that generates natural yet diverse augmented examples by formulating DA as a seq2seq task. The key technical novelty of Rotom is a meta-learning framework that automatically learns a policy for combining examples from different DA operators, whereby combinatorially reduces the hyper-parameters space. Our experimental results show that Rotom effectively improves a model's performance by combining multiple DA operators, even when applying them individually does not yield performance improvement. With this strength, Rotom outperforms the state-of-the-art entity matching and data cleaning systems in the low-resource settings as well as two recently proposed DA techniques for text classification.
This paper introduces a novel evaluation methodology for entity resolution algorithms. It is motivated by PatentsView.org, a U.S. Patents and Trademarks Office patent data exploration tool that disambiguates patent inventors using an entity resolution algorithm. We provide a data collection methodology and tailored performance estimators that account for sampling biases. Our approach is simple, practical and principled -- key characteristics that allow us to paint the first representative picture of PatentsView's disambiguation performance. This approach is used to inform PatentsView's users of the reliability of the data and to allow the comparison of competing disambiguation algorithms.
Entity resolution, a longstanding problem of data cleaning and integration, aims at identifying data records that represent the same real-world entity. Existing approaches treat entity resolution as a universal task, assuming the existence of a single interpretation of a real-world entity and focusing only on finding matched records, separating corresponding from non-corresponding ones, with respect to this single interpretation. However, in real-world scenarios, where entity resolution is part of a more general data project, downstream applications may have varying interpretations of real-world entities relating, for example, to various user needs. In what follows, we introduce the problem of multiple intents entity resolution (MIER), an extension to the universal (single intent) entity resolution task. As a solution, we propose FlexER, utilizing contemporary solutions to universal entity resolution tasks to solve multiple intents entity resolution. FlexER addresses the problem as a multi-label classification problem. It combines intent-based representations of tuple pairs using a multiplex graph representation that serves as an input to a graph neural network (GNN). FlexER learns intent representations and improves the outcome to multiple resolution problems. A large-scale empirical evaluation introduces a new benchmark and, using also two well-known benchmarks, shows that FlexER effectively solves the MIER problem and outperforms the state-of-the-art for a universal entity resolution.
Neural methods have become the de-facto choice for the vast majority of data analysis tasks, and entity alignment (EA) is no exception. Not surprisingly, more than 50 different neural EA methods have been published since 2017. However, surprisingly, an analysis of the differences between neural and non-neural EA methods has been lacking. We bridge this gap by performing an in-depth comparison among five carefully chosen representative state-of-the-art methods from the pre-neural and neural era. We unravel, and consequently mitigate, the inherent deficiencies in the experimental setup utilized for evaluating neural EA methods. To ensure fairness in evaluation, we homogenize the entity matching modules of neural and non-neural methods. Additionally, for the first time, we draw a parallel between EA and record linkage (RL) by empirically showcasing the ability of RL methods to perform EA. Our results indicate that Paris, the state-of-the-art non-neural method, statistically significantly outperforms all the representative state-of-the-art neural methods in terms of both efficacy and efficiency across a wide variety of dataset types and scenarios, and is second only to BERT-INT for a specific scenario of cross-lingual EA. Our findings shed light on the potential problems resulting from an impulsive application of neural methods as a panacea for all data analytics tasks. Overall, our work results in two overarching conclusions: (1) Paris should be used as a baseline in every follow-up work on EA, and (2) neural methods need to be positioned better to showcase their true potential, for which we provide multiple recommendations.
Entity Resolution (ER) links entities that refer to the same real-world entity from different sources. Existing work usually takes pairs of entities as input and judges those pairs independently. However, there is often interdependence between different pairs of ER decisions, e.g., the entities from the same data source are usually semantically related to each other. Furthermore, current ER approaches are mainly based on attribute similarity comparison, but ignore interdependence between attributes. To address the limits of existing methods, we propose HierGAT, a new method for ER based on a Hierarchical Graph Attention Transformer Network, which can model and exploit the interdependence between different ER decisions. The benefit of our method comes from: 1) The graph attention network model for joint ER decisions; 2) The graph-attention capability to identify the discriminative words from attributes and find the most discriminative attributes. Furthermore, we propose to learn contextual embeddings to enrich word embeddings for better performance. The experimental results on publicly available benchmark datasets show that HierGAT outperforms DeepMatcher by up to 32.5% of F1 score and up to 8.7% of F1 score compared with Ditto.
Entity Resolution (ER) in voice assistants is a prime component during run time that resolves entities in users request to real world entities. ER involves two major functionalities 1. Relevance generation and 2. Ranking. In this paper we propose a low cost relevance generation framework by generating features using customer implicit and explicit feedback signals. The generated relevance datasets can serve as test sets to measure ER performance. We also introduce a set of metrics that accurately measures the performance of ER systems in various dimensions. They provide great interpretability to deep dive and identifying root cause of ER issues, whether the problem is in relevance generation or ranking.
Foundation Models (FMs) are models trained on large corpora of data that, at very large scale, can generalize to new tasks without any task-specific finetuning. As these models continue to grow in size, innovations continue to push the boundaries of what these models can do on language and image tasks. This paper aims to understand an underexplored area of FMs: classical data tasks like cleaning and integration. As a proof-of-concept, we cast five data cleaning and integration tasks as prompting tasks and evaluate the performance of FMs on these tasks. We find that large FMs generalize and achieve SoTA performance on data cleaning and integration tasks, even though they are not trained for these data tasks. We identify specific research challenges and opportunities that these models present, including challenges with private and domain specific data, and opportunities to make data management systems more accessible to non-experts. We make our code and experiments publicly available at: https://github.com/HazyResearch/fm_data_tasks.
Entity matching (EM) is the most critical step for entity resolution (ER). While current deep learningbased methods achieve very impressive performance on standard EM benchmarks, their realworld application performance is much frustrating. In this paper, we highlight that such the gap between reality and ideality stems from the unreasonable benchmark construction process, which is inconsistent with the nature of entity matching and therefore leads to biased evaluations of current EM approaches. To this end, we build a new EM corpus and re-construct EM benchmarks to challenge critical assumptions implicit in the previous benchmark construction process by step-wisely changing the restricted entities, balanced labels, and single-modal records in previous benchmarks into open entities, imbalanced labels, and multimodal records in an open environment. Experimental results demonstrate that the assumptions made in the previous benchmark construction process are not coincidental with the open environment, which conceal the main challenges of the task and therefore significantly overestimate the current progress of entity matching. The constructed benchmarks and code are publicly released
Similarity query is the family of queries based on some similarity metrics. Unlike the traditional database queries which are mostly based on value equality, similarity queries aim to find targets "similar enough to" the given data objects, depending on some similarity metric, e.g., Euclidean distance, cosine similarity and so on. To measure the similarity between data objects, traditional methods normally work on low level or syntax features(e.g., basic visual features on images or bag-of-word features of text), which makes them weak to compute the semantic similarities between objects. So for measuring data similarities semantically, neural embedding is applied. Embedding techniques work by representing the raw data objects as vectors (so called "embeddings" or "neural embeddings" since they are mostly generated by neural network models) that expose the hidden semantics of the raw data, based on which embeddings do show outstanding effectiveness on capturing data similarities, making it one of the most widely used and studied techniques in the state-of-the-art similarity query processing research. But there are still many open challenges on the efficiency of embedding based similarity query processing, which are not so well-studied as the effectiveness. In this survey, we first provide an overview of the "similarity query" and "similarity query processing" problems. Then we talk about recent approaches on designing the indexes and operators for highly efficient similarity query processing on top of embeddings (or more generally, high dimensional data). Finally, we investigate the specific solutions with and without using embeddings in selected application domains of similarity queries, including entity resolution and information retrieval. By comparing the solutions, we show how neural embeddings benefit those applications.
Entity resolution (ER) aims at matching records that refer to the same real-world entity. Although widely studied for the last 50 years, ER still represents a challenging data management problem, and several recent works have started to investigate the opportunity of applying deep learning (DL) techniques to solve this problem. In this paper, we study the fundamental problem of explainability of the DL solution for ER. Understanding the matching predictions of an ER solution is indeed crucial to assess the trustworthiness of the DL model and to discover its biases. We treat the DL model as a black box classifier and - while previous approaches to provide explanations for DL predictions are agnostic to the classification task. we propose the CERTA approach that is aware of the semantics of the ER problem. Our approach produces both saliency explanations, which associate each attribute with a saliency score, and counterfactual explanations, which provide examples of values that can flip the prediction. CERTA builds on a probabilistic framework that aims at computing the explanations evaluating the outcomes produced by using perturbed copies of the input records. We experimentally evaluate CERTA's explanations of state-of-the-art ER solutions based on DL models using publicly available datasets, and demonstrate the effectiveness of CERTA over recently proposed methods for this problem.
Contrastive learning has moved the state of the art for many tasks in computer vision and information retrieval in recent years. This poster is the first work that applies supervised contrastive learning to the task of product matching in e-commerce using product offers from different e-shops. More specifically, we employ a supervised contrastive learning technique to pre-train a Transformer encoder which is afterward fine-tuned for the matching task using pair-wise training data. We further propose a source-aware sampling strategy that enables contrastive learning to be applied for use cases in which the training data does not contain product identifiers. We show that applying supervised contrastive pre-training in combination with source-aware sampling significantly improves the state-of-the-art performance on several widely used benchmarks: For Abt-Buy, we reach an F1-score of 94.29 (+3.24 compared to the previous state-of-the-art), for Amazon-Google 79.28 (+ 3.7). For WDC Computers datasets, we reach improvements between +0.8 and +8.84 in F1-score depending on the training set size. Further experiments with data augmentation and self-supervised contrastive pre-training show that the former can be helpful for smaller training sets while the latter leads to a significant decline in performance due to inherent label noise. We thus conclude that contrastive pre-training has a high potential for product matching use cases in which explicit supervision is available.
Random partition models are widely used in Bayesian methods for various clustering tasks, such as mixture models, topic models, and community detection problems. While the number of clusters induced by random partition models has been studied extensively, another important model property regarding the balancedness of partition has been largely neglected. We formulate a framework to define and theoretically study the balancedness of exchangeable random partition models, by analyzing how a model assigns probabilities to partitions with different levels of balancedness. We demonstrate that the "rich-get-richer" characteristic of many existing popular random partition models is an inevitable consequence of two common assumptions: product-form exchangeability and projectivity. We propose a principled way to compare the balancedness of random partition models, which gives a better understanding of what model works better and what doesn't for different applications. We also introduce the "rich-get-poorer" random partition models and illustrate their application to entity resolution tasks.
Entity resolution is a widely studied problem with several proposals to match records across relations. Matching textual content is a widespread task in many applications, such as question answering and search. While recent methods achieve promising results for these two tasks, there is no clear solution for the more general problem of matching textual content and structured data. We introduce a framework that supports this new task in an unsupervised setting for any pair of corpora, being relational tables or text documents. Our method builds a fine-grained graph over the content of the corpora and derives word embeddings to represent the objects to match in a low dimensional space. The learned representation enables effective and efficient matching at different granularity, from relational tuples to text sentences and paragraphs. Our flexible framework can exploit pre-trained resources, but it does not depends on their existence and achieves better quality performance in matching content when the vocabulary is domain specific. We also introduce optimizations in the graph creation process with an "expand and compress" approach that first identifies new valid relationships across elements, to improve matching, and then prunes nodes and edges, to reduce the graph size. Experiments on real use cases and public datasets show that our framework produces embeddings that outperform word embeddings and fine-tuned language models both in results' quality and in execution times.
Here we study the problem of matched record clustering in unsupervised entity resolution. We build upon a state-of-the-art probabilistic framework named the Data Washing Machine (DWM). We introduce a graph-based hierarchical 2-step record clustering method (GDWM) that first identifies large, connected components or, as we call them, soft clusters in the matched record pairs using a graph-based transitive closure algorithm utilized in the DWM. That is followed by breaking down the discovered soft clusters into more precise entity clusters in a hierarchical manner using an adapted graph-based modularity optimization method. Our approach provides several advantages over the original implementation of the DWM, mainly a significant speed-up, increased precision, and overall increased F1 scores. We demonstrate the efficacy of our approach using experiments on multiple synthetic datasets. Our results also provide evidence of the utility of graph theory-based algorithms despite their sparsity in the literature on unsupervised entity resolution.
Multipartite entity resolution aims at integrating records from multiple datasets into one entity. We derive a mathematical formulation for a general class of record linkage problems in multipartite entity resolution across many datasets as a combinatorial optimization problem known as the multidimensional assignment problem. As a motivation for our approach, we illustrate the advantage of multipartite entity resolution over sequential bipartite matching. Because the optimization problem is NP-hard, we apply two heuristic procedures, a Greedy algorithm and very large scale neighborhood search, to solve the assignment problem and find the most likely matching of records from multiple datasets into a single entity. We evaluate and compare the performance of these algorithms and their modifications on synthetically generated data. We perform computational experiments to compare performance of recent heuristic, the very large-scale neighborhood search, with a Greedy algorithm, another heuristic for the MAP, as well as with two versions of genetic algorithm, a general metaheuristic. Importantly, we perform experiments to compare two alternative methods of re-starting the search for the former heuristic, specifically a random-sampling multi-start and a deterministic design-based multi-start. We find evidence that design-based multi-start can be more efficient as the size of databases grow large. In addition, we show that very large scale search, especially its multi-start version, outperforms simple Greedy heuristic. Hybridization of Greedy search with very large scale neighborhood search improves the performance. Using multi-start with as few as three additional runs of very large scale search offers some improvement in the performance of the very large scale search procedure. Last, we propose an approach to evaluating complexity of the very large-scale neighborhood search.
Entity matching is a central task in data integration which has been researched for decades. Over this time, a wide range of benchmark tasks for evaluating entity matching methods has been developed. This resource paper systematically complements, profiles, and compares 21 entity matching benchmark tasks. In order to better understand the specific challenges associated with different tasks, we define a set of profiling dimensions which capture central aspects of the matching tasks. Using these dimensions, we create groups of benchmark tasks having similar characteristics. Afterwards, we assess the difficulty of the tasks in each group by computing baseline evaluation results using standard feature engineering together with two common classification methods. In order to enable the exact reproducibility of evaluation results, matching tasks need to contain exactly defined sets of matching and non-matching record pairs, as well as a fixed development and test split. As this is not the case for some widely-used benchmark tasks, we complement these tasks with fixed sets of non-matching pairs, as well as fixed splits, and provide the resulting development and test sets for public download. By profiling and complementing the benchmark tasks, we support researchers to select challenging as well as diverse tasks and to compare matching systems on clearly defined grounds.
Entity matching (EM) finds data instances that refer to the same real-world entity. In this paper we examine applying deep learning (DL) to EM, to understand DL's benefits and limitations. We review many DL solutions that have been developed for related matching tasks in text processing (e.g., entity linking, textual entailment, etc.). We categorize these solutions and define a space of DL solutions for EM, as embodied by four solutions with varying representational power: SIF, RNN, Attention, and Hybrid. Next, we investigate the types of EM problems for which DL can be helpful. We consider three such problem types, which match structured data instances, textual instances, and dirty instances, respectively. We empirically compare the above four DL solutions with Magellan, a state-of-the-art learning-based EM solution. The results show that DL does not outperform current solutions on structured EM, but it can significantly outperform them on textual and dirty EM. For practitioners, this suggests that they should seriously consider using DL for textual and dirty EM problems. Finally, we analyze DL's performance and discuss future research directions.
An increasing number of data providers have adopted shared numbering schemes such as GTIN, ISBN, DUNS, or ORCID numbers for identifying entities in the respective domain. This means for data integration that shared identifiers are often available for a subset of the entity descriptions to be integrated while such identifiers are not available for others. The challenge in these settings is to learn a matcher for entity descriptions without identifiers using the entity descriptions containing identifiers as training data. The task can be approached by learning a binary classifier which distinguishes pairs of entity descriptions for the same real-world entity from descriptions of different entities. The task can also be modeled as a multi-class classification problem by learning classifiers for identifying descriptions of individual entities. We present a dual-objective training method for BERT, called JointBERT, which combines binary matching and multi-class classification, forcing the model to predict the entity identifier for each entity description in a training pair in addition to the match/non-match decision. Our evaluation across five entity matching benchmark datasets shows that dual-objective training can increase the matching performance for seen products by 1% to 5% F1 compared to single-objective Transformer-based methods, given that enough training data is available for both objectives. In order to gain a deeper understanding of the strengths and weaknesses of the proposed method, we compare JointBERT to several other BERT-based matching methods as well as baseline systems along a set of specific matching challenges. This evaluation shows that JointBERT, given enough training data for both objectives, outperforms the other methods on tasks involving seen products, while it underperforms for unseen products. Using a combination of LIME explanations and domain-specific word classes, we analyze the matching decisions of the different deep learning models and conclude that BERT-based models are better at focusing on relevant word classes compared to RNN-based models.