Entity Resolution Research Articles

Reproducible experiments on Three-Dimensional Entity Resolution with JedAI

In Papadakis et al. (2020), we presented the latest release of JedAI, an open-source Entity Resolution (ER) system that allows for building a large variety of end-to-end ER pipelines. Through a thorough experimental evaluation, we compared a schema-agnostic ER pipeline based on blocks with another schema-based ER pipeline based on similarity joins. We applied them to 10 established, real-world datasets and assessed them with respect to effectiveness and time efficiency. Special care was taken to juxtapose their scalability, too, using seven established, synthetic datasets. Moreover, we experimentally compared the effectiveness of the batch schema-agnostic ER pipeline with its progressive counterpart. In this companion paper, we describe how to reproduce the entire experimental study that pertains to JedAI’s serial execution through its intuitive user interface. We also explain how to examine the robustness of the parameter configurations we have selected.

A budget-limited mechanism for category-aware crowdsourcing of multiple-choice tasks

Crowdsourcing harnesses human effort to solve computer-hard problems such as photo tagging, entity resolution and sentiment analysis. Such tasks often have different levels of difficulty and workers have varying levels of skill at completing them. With a limited budget, it is important to wisely allocate the spend among the tasks and workers such that the overall outcome is as good as possible. Most existing work addresses this budget allocation problem by assuming that workers have a single level of ability for all tasks and each task involves a choice between just two alternatives. However, this neglects the fact that many crowdsourcing applications ask workers to choose between multiple alternatives and that different tasks can belong to a variety of diverse categories. Moreover, workers may have varying abilities across these categories. For example, a science enthusiast is likely to do better than a cinephile when answering a question such as “selecting the melting point of Copper from 1) 327 degrees Celcius, 2) 1085 degrees Celcius and 3) 1495 degrees Celcius”. And a cinephile is likely to perform better in tasks related to movies such as “how many episodes of Fooly Cooly were ever made? 1) 6 2) 7 and 3) 8”. To incorporate such category-aware crowdsourcing of multiple-choice tasks, we model the interaction between the crowdsource campaign initiator and the workers as a procurement auction and propose a computationally efficient mechanism, INCARE, to achieve high-quality outcomes given a limited budget. We prove that INCARE is budget feasible, incentive compatible and individually rational. We also prove that INCARE can achieve a bounded approximation ratio for the optimal budget allocation mechanism with full knowledge of workers' true costs. Finally, our numerical simulations, on both real and synthetic data, show that, compared to the state of the art, INCARE: (i) can improve the accuracy by up to 98%, given a limited budget; and (ii) is significantly more robust to inaccuracies in prior information about each worker's ability and each task's difficulty.

Semantic-Based and Entity-Resolution Fusion to Enhance Quality of Big RDF Data

Within an organisation, the quality in big data is a cornerstone to operational, transactional processes and to the reliability of business analytics for decision making. In fact, as organizations are harnessing multi-sources data to rise the benefits of their business, the quality of data becomes important and crucial. This paper presents a new approach to query big data sources using Resource Description Framework (RDF) representation to ensure data quality by harvesting more relevant and complete query results. Our approach handles two important types of heterogeneity over multiple data sources: semantic heterogeneity and URI-based entity identification. It proposes (1) a semantic entity resolution method based on inference mechanism using rules to manage the misunderstanding of data, in real world entities (2) Data Quality enhancement using MapReduce-based query rewriting approach includes the entity resolution results to infer and adds implicit data into query results (3) a parallel combination of MapReduce jobs of saturation and query rewriting inferences to handle transitive and cyclic rules for a richer rules' expression language (4) experiments to assess the efficiency of the proposed approach over real big RDF data originating from insurance and synthetic data sets.

Improving the Efficiency and Effectiveness for BERT-based Entity Resolution

BERT has set a new state-of-the-art performance on entity resolution (ER) task, largely owed to fine-tuning pre-trained language models and the deep pair-wise interaction. Albeit being remarkably effective, it comes with a steep increase in computational cost, as the deep-interaction requires to exhaustively compute every tuple pair to search for co-references. For ER task, it is often prohibitively expensive due to the large cardinality to be matched. To tackle this, we introduce a siamese network structure that independently encodes tuples using BERT but delays the pair-wise interaction via an enhanced alignment network. This siamese structure enables a dedicated blocking module to quickly filter out obviously dissimilar tuple pairs, and thus drastically reduces the cardinality of fine-grained matching. Further, the blocking and entity matching are integrated into a multi-task learning framework for facilitating both tasks. Extensive experiments on multiple datasets demonstrate that our model significantly outperforms state-of-the-art models (including BERT) in both efficiency and effectiveness.

Attention-Enhanced Gradual Machine Learning for Entity Resolution

Recent work has shown that entity resolution (ER) can be effectively performed by gradual machine learning (GML). GML begins with some automatically labeled easy instances and, then, gradually labels more challenging instances by iterative factor graph inference without human intervention. In GML, shared features serve as the medium for knowledge conveyance between easy instances and more challenging ones. The existing GML solution supposes that features play independent roles in gradual inference. However, in real scenarios, this assumption may be untenable since features are usually correlated with each other. To address this limitation, this article proposes an attention-enhanced approach to improve the accuracy of gradual inference. We first propose a method of spectral feature representation to map correlated features to close points in the same vector space and, then, present a model of attention neural network to learn the decisive features, given arbitrary combinations of features for improved feature weighting. Finally, our extensive experiments on real benchmark data have validated the efficacy of the proposed approach.

HySet: A hybrid framework for exact set similarity join using a GPU

Set similarity join is a fundamental operation used in a wide range of applications such as data mining, data cleaning and entity resolution. Existing methods proposed for set similarity join conform to a filter-verification framework where potential candidate pairs are generated in the filtering phase and then undergo a verification phase to output the final result. Several different kinds of filtering techniques have been proposed and techniques also differentiate in the manner they couple filtering with verification. However, it has been shown that no globally dominant technique exists. Depending on the dataset and query characteristics, each technique has its own strong and weak points. Based on these findings, the main contribution of this work is the development of a hybrid framework for the set similarity join operation for a single GPU-equipped machine setting. Our framework encapsulates a partitioning mechanism to utilize appropriately both the CPU and the GPU. We present all technical details and we show performance speedups up to 3.25x after thorough evaluation.

RPT

Can AI help automate human-easy but computer-hard data preparation tasks that burden data scientists, practitioners, and crowd workers? We answer this question by presenting RPT, a denoising autoencoder for tuple-to-X models (" X " could be tuple, token, label, JSON, and so on). RPT is pre-trained for a tuple-to-tuple model by corrupting the input tuple and then learning a model to reconstruct the original tuple. It adopts a Transformer-based neural translation architecture that consists of a bidirectional encoder (similar to BERT) and a left-to-right autoregressive decoder (similar to GPT), leading to a generalization of both BERT and GPT. The pre-trained RPT can already support several common data preparation tasks such as data cleaning, auto-completion and schema matching. Better still, RPT can be fine-tuned on a wide range of data preparation tasks, such as value normalization, data transformation, data annotation, etc. To complement RPT, we also discuss several appealing techniques such as collaborative training and few-shot learning for entity resolution, and few-shot learning and NLP question-answering for information extraction. In addition, we identify a series of research opportunities to advance the field of data preparation.

Parallel discrepancy detection and incremental detection

This paper studies how to catch duplicates, mismatches and conflicts in the same process. We adopt a class of entity enhancing rules that embed machine learning predicates, unify entity resolution and conflict resolution, and are collectively defined across multiple relations. We detect discrepancies as violations of such rules. We establish the complexity of discrepancy detection and incremental detection problems with the rules; they are both NP-complete and W[1]-hard. To cope with the intractability and scale with large datasets, we develop parallel algorithms and parallel incremental algorithms for discrepancy detection. We show that both algorithms are parallelly scalable, i.e. , they guarantee to reduce runtime when more processors are used. Moreover, the parallel incremental algorithm is relatively bounded. The complexity bounds and algorithms carry over to denial constraints, a special case of the entity enhancing rules. Using real-life and synthetic datasets, we experimentally verify the effectiveness, scalability and efficiency of the algorithms.

Features Selection for Entity Resolution in Prostitution on Twitter

Entity resolution is the process of determining whether two references to real-world objects refer to the same or different purposes. This study applies entity resolution on Twitter prostitution dataset based on features with the Regularized Logistic Regression training and determination of Active Learning on Dedupe and based on graphs using Neo4j and Node2Vec. This study found that maximum similarity is 1 when the number of features (personal, location and bio specifications) is complete. The minimum similarity is 0.025662627 when the amount of harmful training data. The most influencing similarity feature is the cellphone number with the lowest starting range from 0.997678459 to 0.999993523. The parameter - length of walk per source has the effect of achieving the best similarity accuracy reaching 71.4% (prediction 14 and yield 10).

Unsupervised DNF Blocking for Efficient Linking of Knowledge Graphs and Tables

Entity Resolution (ER) is the problem of identifying co-referent entity pairs across datasets, including knowledge graphs (KGs). ER is an important prerequisite in many applied KG search and analytics pipelines, with a typical workflow comprising two steps. In the first ’blocking’ step, entities are mapped to blocks. Blocking is necessary for preempting comparing all possible pairs of entities, as (in the second ‘similarity’ step) only entities within blocks are paired and compared, allowing for significant computational savings with a minimal loss of performance. Unfortunately, learning a blocking scheme in an unsupervised fashion is a non-trivial problem, and it has not been properly explored for heterogeneous, semi-structured datasets, such as are prevalent in industrial and Web applications. This article presents an unsupervised algorithmic pipeline for learning Disjunctive Normal Form (DNF) blocking schemes on KGs, as well as structurally heterogeneous tables that may not share a common schema. We evaluate the approach on six real-world dataset pairs, and show that it is competitive with supervised and semi-supervised baselines.

A Multi-Disciplinary Perspective for Conducting Artificial Intelligence-enabled Privacy Analytics

Events such as Facebook-Cambridge Analytica scandal and data aggregation efforts by technology providers have illustrated how fragile modern society is to privacy violations. Internationally recognized entities such as the National Science Foundation (NSF) have indicated that Artificial Intelligence (AI)-enabled models, artifacts, and systems can efficiently and effectively sift through large quantities of data from legal documents, social media, Dark Web sites, and other sources to curb privacy violations. Yet considerable efforts are still required for understanding prevailing data sources, systematically developing AI-enabled privacy analytics to tackle emerging challenges, and deploying systems to address critical privacy needs. To this end, we provide an overview of prevailing data sources that can support AI-enabled privacy analytics; a multi-disciplinary research framework that connects data, algorithms, and systems to tackle emerging AI-enabled privacy analytics challenges such as entity resolution, privacy assistance systems, privacy risk modeling, and more; a summary of selected funding sources to support high-impact privacy analytics research; and an overview of prevailing conference and journal venues that can be leveraged to share and archive privacy analytics research. We conclude this paper with an introduction of the papers included in this special issue.

SDLER: stacked dedupe learning for entity resolution in big data era

In the Big Data Era, Entity Resolution (ER) faces many challenges such as high scalability, the coexistence of complex similarity metrics, tautonymy and synonym, and the requirement of Data Quality Evaluation. Moreover, despite more than seventy years of development efforts, there is still a high demand for democratizing ER to reduce human participation in tuning parameters, data labeling, defining blocking functions, and feature engineering. This study aimed to explore a novel Stacked Dedupe Learning ER system with high accuracy and efficiency. The study evaluated sophisticated composition methods, such as Bidirectional Recurrent Neural Networks (BiRNNs) and Long Short-Term Memory (LSTM) hidden units, to renovate each tuple to word representation distribution in a sense to capture similarities amidst tuples. Also, pre-trained words embedding where they were not available, ways to learn and tune Word Representation Distribution customized for ER tasks under different scenarios were considered. More so, the Locality Sensitive Hashing (LSH) based blocking approach, which considered the entire attributes of a tuple and produced slighter blocks, compared with traditional methods with few attributes, were assessed. The algorithm was tested on multiple datasets namely benchmarks, and multi-lingual data. The experimental results showed that Stacked Dedupe Learning achieves high quality and good performance, and scales well compared to the existing solutions.

Efficient and effective ER with progressive blocking

Blocking is a mechanism to improve the efficiency of Entity Resolution (ER) which aims to quickly prune out all non-matching record pairs. However, depending on the distributions of entity cluster sizes, existing techniques can be either (a) too aggressive, such that they help scale but can adversely affect the ER effectiveness, or (b) too permissive, potentially harming ER efficiency. In this paper, we propose a new methodology of progressive blocking (pBlocking) to enable both efficient and effective ER, which works seamlessly across different entity cluster size distributions. pBlocking is based on the insight that the effectiveness-efficiency trade-off is revealed only when the output of ER starts to be available. Hence, pBlocking leverages partial ER output in a feedback loop to refine the blocking result in a data-driven fashion. Specifically, we bootstrap pBlocking with traditional blocking methods and progressively improve the building and scoring of blocks until we get the desired trade-off, leveraging a limited amount of ER results as a guidance at every round. We formally prove that pBlocking converges efficiently ($O(n log^2 n)$ time complexity, where n is the total number of records). Our experiments show that incorporating partial ER output in a feedback loop can improve the efficiency and effectiveness of blocking by 5x and 60% respectively, improving the overall F-score of the entire ER process up to 60%.

Lost and found: applying network analysis to public health contact tracing for HIV

Infectious disease surveillance is often case-based, focused on people diagnosed and their contacts in a predefined time window, and treated as independent across infections. Network analysis of partners and contacts joining multiple investigations and infections can reveal social or temporal trends, providing opportunities for epidemic control within broader networks. We constructed a sociosexual network of all HIV and early syphilis cases and contacts investigated among residents of 11 contiguous counties in North Carolina over a two-year period (2012–2013). We anchored the analysis on new HIV diagnoses (“indexes”), but also included nodes and edges from syphilis investigations that were within the same network component as any new HIV index. After adding syphilis investigations and deduplicating people included in multiple investigations (entity resolution), the final network comprised 1470 people: 569 HIV indexes, 700 contacts to HIV indexes who were not also new cases themselves, and 201 people who were either indexes or contacts in eligible syphilis investigations. Among HIV indexes, nearly half (48%; n = 273) had no located contacts during single-investigation contact tracing, though 25 (9%) of these were identified by other network members and thus not isolated in the final multiple investigation network. Constructing a sociosexual network from cases and contacts across multiple investigations mitigated some effects of unobserved partnerships underlying the HIV epidemic and demonstrated the HIV and syphilis overlap in these networks.

D-blink: Distributed End-to-End Bayesian Entity Resolution

Entity resolution (ER; also known as record linkage or de-duplication) is the process of merging noisy databases, often in the absence of unique identifiers. A major advancement in ER methodology has been the application of Bayesian generative models, which provide a natural framework for inferring latent entities with rigorous quantification of uncertainty. Despite these advantages, existing models are severely limited in practice, as standard inference algorithms scale quadratically in the number of records. While scaling can be managed by fitting the model on separate blocks of the data, such a naïve approach may induce significant error in the posterior. In this article, we propose a principled model for scalable Bayesian ER, called “distributed Bayesian linkage” or d-blink, which jointly performs blocking and ER without compromising posterior correctness. Our approach relies on several key ideas, including: (i) an auxiliary variable representation that induces a partition of the entities and records into blocks; (ii) a method for constructing well-balanced blocks based on k-d trees; (iii) a distributed partially collapsed Gibbs sampler with improved mixing; and (iv) fast algorithms for performing Gibbs updates. Empirical studies on six datasets—including a case study on the 2010 Decennial Census—demonstrate the scalability and effectiveness of our approach. Supplementary materials for this article are available online.

Information networks fusion based on multi-task coordination

Information networks provide a powerful representation of entities and the relationships between them. Information networks fusion is a technique for information fusion that jointly reasons about entities, links and relations in the presence of various sources. However, existing methods for information networks fusion tend to rely on a single task which might not get enough evidence for reasoning. In order to solve this issue, in this paper, we present a novel model called MC-INFM (information networks fusion model based on multi-task coordination). Different from traditional models, MC-INFM casts the fusion problem as a probabilistic inference problem, and collectively performs multiple tasks (including entity resolution, link prediction and relation matching) to infer the final result of fusion. First, we define the intra-features and the inter-features respectively and model them as factor graphs, which can provide abundant evidence to infer. Then, we use conditional random field (CRF) to learn the weight of each feature and infer the results of these tasks simultaneously by performing the maximum probabilistic inference. Experiments demonstrate the effectiveness of our proposed model.

A novel similarity measure for spatial entity resolution based on data granularity model: Managing inconsistencies in place descriptions

Tremendous amounts of data are generated every day by different sources and stored in heterogeneous databases. Providing an integrated view by fusion of data is essential to enhance data utilization. An indispensable type of data is spatial data, with diverse application domains, including GIS, e-commerce, military, and tourism. The concept of location forms a key part of user-generated data with serious challenges, including uncertainty. A particular location may have different names, and conversely, various locations may have the same name. Furthermore, geographical coordinates of locations may not be expressed accurately in datasets. More challenges also exist that have received less attention. Various data sources might describe locations in different levels of detail. This increases data inconsistency and decreases the quality of data fusion. This paper focuses on spatial data granulation to deal with this variety. If these diversities are not taken into consideration, the different descriptions of a location may be interpreted differently and, in turn, not be fused. The contribution of this paper are: (a) Introducing a granular approach to measure the similarity between two place description for managing apparent differences. The proposed method improves the quality of the geocoding and data fusion phases, (b) Introducing a novel data blocking method to decrease pairwise comparisons based on geographical features. For result evaluation, we developed a dataset from two real aviation accident datasets. The evaluation shows that the quality of entity recognition and data fusion improved by using our proposed data granulation technique.

Knowledge Transfer for Entity Resolution with Siamese Neural Networks

The integration of multiple data sources is a common problem in a large variety of applications. Traditionally, handcrafted similarity measures are used to discover, merge, and integrate multiple representations of the same entity—duplicates—into a large homogeneous collection of data. Often, these similarity measures do not cope well with the heterogeneity of the underlying dataset. In addition, domain experts are needed to manually design and configure such measures, which is both time-consuming and requires extensive domain expertise. We propose a deep Siamese neural network, capable of learning a similarity measure that is tailored to the characteristics of a particular dataset. With the properties of deep learning methods, we are able to eliminate the manual feature engineering process and thus considerably reduce the effort required for model construction. In addition, we show that it is possible to transfer knowledge acquired during the deduplication of one dataset to another, and thus significantly reduce the amount of data required to train a similarity measure. We evaluated our method on multiple datasets and compare our approach to state-of-the-art deduplication methods. Our approach outperforms competitors by up to +26 percent F-measure, depending on task and dataset. In addition, we show that knowledge transfer is not only feasible, but in our experiments led to an improvement in F-measure of up to +4.7 percent.

Generating Name-Like Vectors for Testing Large-Scale Entity Resolution

Entity resolution (ER), the problem of identifying and linking records that belong to the same real-world entities in structured and unstructured data, is a primary task in data integration. Accurate and efficient ER has a major practical impact on various applications across commercial, security and scientific domains. Recently, scalable ER techniques have received enormous attention with the increasing need to combine large-scale datasets. The shortage of training and ground truth data impedes the development and testing of ER algorithms. Good public datasets, especially those containing personal information, are restricted in this area and usually small in size. Due to privacy and confidential issues, testing algorithms or techniques with real datasets is challenging in ER research. Simulation is one technique for generating synthetic datasets that have characteristics similar to those of real data for testing algorithms. Many existing simulation tools in ER lack support for generating large-scale data and have problems in complexity, scalability, and limitations of resampling. In our work, we propose a simple, inexpensive, and fast synthetic data generation tool. Our tool only generates entity names in the first stage, but these are commonly used as identification keys in ER algorithms. We avoid the detail-level simulation of entity names using a simple vector representation that delivers simplicity and efficiency. In this paper, we discuss how to simulate simple vectors that approximate the properties of entity names. We describe the overall construction of the tool based on data analysis of a namespace that contains entity names collected from the actual environment.

PSOBER: PSO based entity resolution

Entity Resolution is the task of mapping the records within a database to their corresponding entities. The entity resolution problem presents a lot of challenges because of the absence of complete information in records, variant distribution of records for different entities and sometimes overlaps between records of different entities. In this paper, we have proposed an unsupervised method to solve this problem. The previously mentioned problem is set as a partitioning problem. Thereafter, an optimization algorithm-based technique is proposed to solve the entity resolution problem. The presented approach enables the partitioning of records across entities. A comparative analysis with the genetic algorithm over datasets proves the efficiency of the considered approach.