A Bayesian Approach to Discovering Truth from Conflicting Sources for Data Integration

By Bo Zhao, Benjamin I. P. Rubinstein, J. Gemmell, Jiawei Han. 2012

355 citations recorded by the source corpus.

In practical data integration systems, it is common for the data sources being integrated to provide conflicting information about the same entity. Consequently, a major challenge for data integration is to derive the most complete and accurate integrated records from diverse and sometimes conflicting sources. We term this challenge the truth finding problem. We observe that some sources are generally more reliable than others, and therefore a good model of source quality is the key to solving the truth finding problem. In this work, we propose a probabilistic graphical model that can automatically infer true records and source quality without any supervision. In contrast to previous methods, our principled approach leverages a generative process of two types of errors (false positive and false negative) by modeling two different aspects of source quality. In so doing, ours is also the first approach designed to merge multi-valued attribute types. Our method is scalable, due to an efficient sampling-based inference algorithm that needs very few iterations in practice and enjoys linear time complexity, with an even faster incremental variant. Experiments on two real world datasets show that our new method outperforms existing state-of-the-art approaches to the truth finding problem.

Subfields: Artificial Intelligence, Information Systems, Management Science and Operations Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

In Management Science and Operations Research

Citations in the corpus

A Survey on Truth Discovery (2015, methodology): “In the light of this challenge, the topic of truth discovery [13,19,20,25–27,32,34,53,56,57] has gained increasing popularity recently due to its ability to estimate source reliability degrees and infer true information.”
Data X-Ray: A Diagnostic Tool for Data Errors (2015, background): “Data management research has supplied a variety of tools to deal with errors originating from integrating data from multiple sources [1,5,49,52], identifying data items that refer to the same entity [34,39], resolving conflicts [22, 62, 63], and providing language extensions for cleaning [28].”
Resolving conflicts in heterogeneous data by truth discovery and source reliability estimation (2014, neutral): “Existing truth discovery approaches [3, 8, 9, 11, 12, 14–16, 18, 21–24] and the proposed method try to estimate source reliability in an unsupervised manner.”
Truth Finding on the Deep Web: Is the Problem Solved? (2012, background): “Existing works on data fusion ([3, 8] as surveys and [10, 13, 14, 17, 19, 20] as recent works) have experimented on data collected from the Web in domains such as book, restaurant and sports.”
More Is Better: Recent Progress in Multi-Omics Data Integration Methods (2017, methodology): “Bayesian methods have been applied to data integration for over a decade (Imoto et al., 2004; Zhao et al., 2012).”
Big data integration (2013, background): “To address such veracity related challenges, techniques have been proposed to find the single truth from conflicting values [49], [32], [48], [39], [38], [18], [9], [47] and to find multiple truths [50].”
Snorkel DryBell: A Case Study in Deploying Weak Supervision at Industrial Scale (2018, methodology): “, the latent truth model [31], use generative models similar to the one in Snorkel DryBell in that they represent the unobserved truth as a latent variable.”
Fusing data with correlations (2014, methodology): “The model closest to ours is LTM [25]; we have made detailed comparisons in Section 3 and in experiments.”
A Confidence-Aware Approach for Truth Discovery on Long-Tail Data (2014, methodology): “To overcome this limitation, techniques have been proposed to simultaneously derive trustworthy facts and estimate source reliability degrees [8, 9, 16, 18–20, 23–26, 31, 34–37].”
Snorkel: rapid training data creation with weak supervision (2019, background): “Many approaches to these problems use probablistic graphical models that are related to Snorkel’s generative model in that they represent the unobserved truth as a latent variable, e.g., the latent truth model [54].”
Mining heterogeneous information networks: a structural analysis approach (2013, background): “LTM [30], can be developed to resolve the conflicting source problem effectively.”
Language-Aware Truth Assessment of Fact Candidates (2014, background): “There is a fairly small body of work on truth-ﬁnding (Yin et al., 2007; Galland et al., 2010; Pasternack and Roth, 2010; Li et al., 2011; Yin and Tan, 2011; Zhao et al., 2012; Pasternack and Roth, 2013).”
SLiMFast: Guaranteed Results for Data Fusion and Source Reliability (2015, background): “In particular, certain data fusion models [9, 43] rely on Naive Bayes and assume that source observations are conditionally independent.”