Collective entity resolution in relational data

Journal of the Association for Information Science and Technology, 2013

Log analysis shows that PubMed users frequently use author names in queries for retrieving scientific literature. However, author name ambiguity may lead to irrelevant retrieval results. To improve the PubMed user experience with author name queries, we designed an author name disambiguation system consisting of similarity estimation and agglomerative clustering. A machine-learning method was employed to score the features for disambiguating a pair of papers with ambiguous names. These features enable the computation of pairwise similarity scores to estimate the probability of a pair of papers belonging to the same author, which drives an agglomerative clustering algorithm regulated by 2 factors: name compatibility and probability level. With transitivity violation correction, high precision author clustering is achieved by focusing on minimizing false-positive pairing. Disambiguation performance is evaluated with manual verification of random samples of pairs from clustering results. When compared with a state-of-the-art system, our evaluation shows that among all the pairs the lumping error rate drops from 10.1% to 2.2% for our system, while the splitting error rises from 1.8% to 7.7%. This results in an overall error rate of 9.9%, compared with 11.9% for the state-of-the-art method. Other evaluations based on gold standard data also show the increase in accuracy of our clustering. We attribute the performance improvement to the machine-learning method driven by a largescale training set and the clustering algorithm regulated by a name compatibility scheme preferring precision. With integration of the author name disambiguation system into the PubMed search engine, the overall clickthrough-rate of PubMed users on author name query results improved from 34.9% to 36.9%.

Scientometrics, 2014

We describe ongoing research where the aim is to apply recent results from the research field of information fusion to bibliometric analysis and information retrieval. We highlight the importance of 'uncertainty' within information fusion and argue that this concept is crucial also for bibliometrics and information retrieval. More specifically, we elaborate on three research strategies related to uncertainty: uncertainty management methods, explanation of uncertainty and visualization of uncertainty. We exemplify our strategies to the classical problem of author name disambiguation where we show how uncertainty can be modeled explained and visualized using information fusion. We show how an information seeker can benefit from tracing increases/decreases of uncertainty in the reasoning process and how such changes can be explained through visualization techniques, all for the purpose of that the information seeker should comprehend the complexity involved in the process of modeling and managing uncertainty in bibliometric analysis. Finally we argue that a further integration of information fusion approaches in the research area of bibliometrics and information retrieval may results in new and fruitful venues of research.

Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

How can we help an investigator to efficiently connect the dots and uncover the network of individuals involved in a criminal activity based on the evidence of their connections, such as visiting the same address, or transacting with the same bank account? We formulate this problem as Active Search of Connections, which finds target entities that share evidence of different types with a given lead, where their relevance to the case is queried interactively from the investigator. We present RedThread, an efficient solution for inferring related and relevant nodes while incorporating the user's feedback to guide the inference. Our experiments focus on case building for combating human trafficking, where the investigator follows leads to expose organized activities, i.e. different escort advertisements that are connected and possibly orchestrated. RedThread is a local algorithm and enables online case building when mining millions of ads posted in one of the largest classified advertising websites. The results of RedThread are interpretable, as they explain how the results are connected to the initial lead. We experimentally show that RedThread learns the importance of the different types and different pieces of evidence, while the former could be transferred between cases.

Lecture Notes in Computer Science, 2011

Data Mining and Knowledge Discovery

The extraction and understanding of temporal events and their relations are major challenges in natural language processing. Processing text on a sentence-by-sentence or expression-by-expression basis often fails, in part due to the challenge of capturing the global consistency of the text. We present an ensemble method, which reconciles the outputs of multiple classifiers of temporal expressions across the text using integer programming. Computational experiments show that the ensemble improves upon the best individual results from two recent challenges, SemEval-2013 TempEval-3 (Temporal Annotation) and SemEval-2016 Task 12 (Clinical TempEval).

Pacific-Asia Conference on Knowledge Discovery and Data Mining, 2021

The limited analytical value of using individual databases on their own increasingly requires the integration of large and complex databases for advanced data analytics. Linking personal medical records with travel and immigration data, for example, will allow the effective management of pandemics such as the current COVID-19 outbreak by tracking potentially infected individuals and their contacts. One major challenge for accurate linkage of large databases is the quadratic or even higher computational complexities of many advanced linkage algorithms. In this paper we present a novel approach that, based on the expected number of true matches between two databases, applies active learning to remove compared record pairs that are likely non-matches before a computationally expensive classification or clustering algorithm is employed to classify record pairs. Unlike blocking and indexing techniques that are used to reduce the number of record pairs to be compared, using recursive binning on a data dimension such as time or space, our approach removes likely non-matching record pairs in each bin after their comparison. Experiments on two real-world databases show that similarity filtering can substantially reduce run time and improve precision, at the costs of a small reduction in recall, of the final linkage results.

2018 IEEE 34th International Conference on Data Engineering (ICDE), 2018

Entity Resolution (ER) aims to identify different descriptions in various Knowledge Bases (KBs) that refer to the same entity. ER is challenged by the Variety, Volume and Veracity of descriptions published in the Web of Data. To address them, we propose the MinoanER framework that fulfills full automation and support of highly heterogeneous entities. MinoanER leverages a token-based similarity of entities to define a new metric that derives the similarity of neighboring entities from the most important relations, indicated only by statistics. For high efficiency, similarities are computed from a set of schema-agnostic blocks and processed in a non-iterative way that involves four threshold-free heuristics. We demonstrate that the effectiveness of MinoanER is comparable to existing ER tools over real KBs exhibiting low heterogeneity in terms of entity types and content. Yet, MinoanER outperforms state-of-the-art ER tools when matching highly heterogeneous KBs.

ACM Computing Surveys

One of the most critical tasks for improving data quality and increasing the reliability of data analytics is Entity Resolution (ER), which aims to identify different descriptions that refer to the same real-world entity. Despite several decades of research, ER remains a challenging problem. In this survey, we highlight the novel aspects of resolving Big Data entities when we should satisfy more than one of the Big Data characteristics simultaneously (i.e., Volume and Velocity with Variety). We present the basic concepts, processing steps, and execution strategies that have been proposed by database, semantic Web, and machine learning communities in order to cope with the loose structuredness , extreme diversity , high speed, and large scale of entity descriptions used by real-world applications. We provide an end-to-end view of ER workflows for Big Data, critically review the pros and cons of existing methods, and conclude with the main open research directions.

Signal and Data Processing

Eastern Economic Journal, 2010

A major emerging problem among consumer finance institutions is that customers that are not well recognized might be riskier than customers that are fully recognized. Fortunately, financial institutions count with external vendors databases that indicate the level of recognition of their customers. However, this information is normally presented as features with partial scores that must be aggregated into an overall matching accuracy score. This score indicates how similar a record is to a master database that contains the best available public information about a specific customer. Additionally, information management and risk management departments of financial institutions may have very different models. Hence, it is necessary to connect the customer recognition information with risk models. This paper studies this problem in two parts: 1). generation of a matching accuracy score to quantify the status of entity resolution between consumer records of a major financial company and an external database, and 2). evaluation of the relationship between the matching accuracy score and several risk segments. As a final result, an overall matching accuracy score is obtained for every customer using the most current account information and a learning algorithm. The matching accuracy score is an indicator of the level of customer recognition. This matching accuracy score is correlated with a risk score (FICO) 1 .

Lecture Notes in Computer Science, 2011

News sources are reliably unreliable. Different news sources may provide significantly differing reports about the same event. Often times, even the same news source may provide widely varying data over a period of time about the same event. Past work on inconsistency management and paraconsistent logics assume that we have "clean" definitions of inconsistency. However, when reasoning about events reported in the news, we need to deal with two unique problems: (i) are two events being reported on the same or are they different? and (ii) what does it mean for two event descriptions to be mutually inconsistent, given that these events are often described using linguistic terms that do not always have a uniquely accepted formal semantics? The answers to these two questions turn out to be closely interlinked. In this paper, we propose a probabilistic logic programming language called PLINI (Probabilistic Logic for Inconsistent News Information) within which users can write rules specifying what they mean by inconsistency in situation (ii) above. We show that PLINI rules can be learned automatically from training data using standard machine learning algorithms. PLINI is a variant of the well known generalized annotated program framework that accounts for similarity of numeric, temporal, and spatial terms occurring in news. We develop a syntax, model theoretic semantics, and fixpoint semantics for PLINI rules, and show how PLINI rules can be used to detect inconsistent news reports.

Proceedings of the 41st ACM SIGMOD-SIGACT-SIGAI Symposium on Principles of Database Systems

In this paper, we revisit the problem of entity resolution and propose a novel, logical framework, LACE, which mixes declarative and procedural elements to achieve a number of desirable properties. Our approach is fundamentally declarative in nature: it utilizes hard and soft rules to specify conditions under which pairs of entity references must or may be merged, together with denial constraints that enforce consistency of the resulting instance. Importantly, however, rule bodies are evaluated on the instance resulting from applying the already 'derived' merges. It is the dynamic nature of our semantics that enables us to capture collective entity resolution scenarios, where merges can trigger further merges, while at the same time ensuring that every merge can be justified. As the denial constraints restrict which merges can be performed together, we obtain a space of (maximal) solutions, from which we can naturally define notions of certain and possible merges and query answers. We explore the computational properties of our framework and determine the precise computational complexity of the relevant decision problems. Furthermore, as a first step towards implementing our approach, we demonstrate how we can encode the various reasoning tasks using answer set programming.

PloS one, 2016

Data from different agencies share data of the same individuals. Linking these datasets to identify all the records belonging to the same individuals is a crucial and challenging problem, especially given the large volumes of data. A large number of available algorithms for record linkage are prone to either time inefficiency or low-accuracy in finding matches and non-matches among the records. In this paper we propose efficient as well as reliable sequential and parallel algorithms for the record linkage problem employing hierarchical clustering methods. We employ complete linkage hierarchical clustering algorithms to address this problem. In addition to hierarchical clustering, we also use two other techniques: elimination of duplicate records and blocking. Our algorithms use sorting as a sub-routine to identify identical copies of records. We have tested our algorithms on datasets with millions of synthetic records. Experimental results show that our algorithms achieve nearly 100...

Techniques and Applications

In the wake of recent terrorist atrocities, intelligence experts have commented that failures in detecting terrorist and criminal activities are not so much due to a lack of data, as they are due to difficulties in relating and interpreting the available intelligence. An intelligent tool for monitoring and interpreting intelligence data will provide a helpful means for intelligence analysts to consider emerging scenarios of plausible threats, thereby offering useful assistance in devising and deploying preventive measures against such possibilities. One of the major problems in need of such attention is detecting false identity that has become the common denominator of all serious crime, especially terrorism. Typical approaches to this problem rely on the similarity measure of textual and other content-based characteristics, which are usually not applicable in the case of deceptive and erroneous description. This barrier may be overcome through link information presented in communic...

Knowledge and Information Systems, 2018

Entity resolution in settings with rich relational structure often introduces complex dependencies between co-references. Exploiting these dependencies is challenging-it requires seamlessly combining statistical, relational, and logical dependencies. One task of particular interest is entity resolution in familial networks. In this setting, multiple partial representations of a family tree are provided, from the perspective of different family members, and the challenge is to reconstruct a family tree from these multiple, noisy, partial views. This reconstruction is crucial for applications such as understanding genetic inheritance, tracking disease contagion, and performing census surveys. Here, we design a model that incorporates statistical signals (such as name similarity), relational information (such as sibling overlap), logical constraints (such as transitivity and bijective matching), and predictions from other algorithms (such as logistic regression and support vector machines), in a collective model. We show how to integrate these features using probabilistic soft logic, a scalable probabilistic programming framework. In experiments on real-world data, our model significantly outperforms state-of-the-art classifiers that use relational features but are incapable of collective reasoning. Rules to Leverage Existing Classification Algorithms SameLR(m 1 , m 2) ⇒ Same(m 1 , m 2) ¬SameLR(m 1 , m 2) ⇒ ¬Same(m 1 , m 2) SameSVMs(m 1 , m 2) ⇒ Same(m 1 , m 2) ¬SameSVMs(m 1 , m 2) ⇒ ¬Same(m 1 , m 2) SameLMTs(m 1 , m 2) ⇒ Same(m 1 , m 2) ¬SameLMTs(m 1 , m 2) ⇒ ¬Same(m 1 , m 2) SameLR(m 1 , m 2) ∧ SameSVMs(m 1 , m 2) ∧ SameLMTs(m 1 , m 2) ⇒ Same(m 1 , m 2) ¬SameLR(m 1 , m 2) ∧ ¬SameSVMs(m 1 , m 2) ∧ ¬SameLMTs(m 1 , m 2) ⇒ ¬Same(m 1 , m 2) Prior Rule ¬Same(m 1 , m 2)

ACM SIGMOD Record, 2012

Name ambiguity in the context of bibliographic citation records is a hard problem that affects the quality of services and content in digital libraries and similar systems. The challenges of dealing with author name ambiguity have led to a myriad of disambiguation methods. Generally speaking, the proposed methods usually attempt to group citation records of a same author by finding some similarity among them or try to directly assign them to their respective authors. Both approaches may either exploit supervised or unsupervised techniques. In this article, we propose a taxonomy for characterizing the current author name disambiguation methods described in the literature, present a brief survey of the most representative ones and discuss several open challenges.

ACM Transactions on Knowledge Discovery from Data

Entity resolution (ER) is the process of linking records that refer to the same entity. Traditionally, this process compares attribute values of records to calculate similarities and then classifies pairs of records as referring to the same entity or not based on these similarities. Recently developed graph-based ER approaches combine relationships between records with attribute similarities to improve linkage quality. Most of these approaches only consider databases containing basic entities that have static attribute values and static relationships, such as publications in bibliographic databases. In contrast, temporal record linkage addresses the problem where attribute values of entities can change over time. However, neither existing graph-based ER nor temporal record linkage can achieve high linkage quality on databases with complex entities , where an entity (such as a person) can change its attribute values over time while having different relationships with other entities a...

Proceedings of the 25th ACM International on Conference on Information and Knowledge Management, 2016

We study the problem of using the crowd to perform entity resolution (ER) on a set of records. For many types of records, especially those involving images, such a task can be difficult for machines, but relatively easy for humans. Typical crowd-based ER approaches ask workers for pairwise judgments between records, which quickly becomes prohibitively expensive even for moderate numbers of records. In this paper, we reduce the cost of pairwise crowd ER approaches by soliciting the crowd for attribute labels on records, and then asking for pairwise judgments only between records with similar sets of attribute labels. However, due to errors induced by crowd-based attribute labeling, a naive attribute-based approach becomes extremely inaccurate even with few attributes. To combat these errors, we use error mitigation strategies which allow us to control the accuracy of our results while maintaining significant cost reductions. We develop a probabilistic model which allows us to determine the optimal, lowest-cost combination of error mitigation strategies needed to achieve a minimum desired accuracy. We test our approach with actual crowdworkers on a dataset of celebrity images, and find that our results yield crowd ER strategies which achieve high accuracy yet are significantly lower cost than pairwise-only approaches.

Domain-Specific Knowledge Graph Construction, 2019

Information Extraction 2.1 Introduction Information extraction (IE) is a fundamental component in any knowledge graph construction pipeline, whether domain-specific or generic [111, 119]. As the name implicitly suggests, the goal of an IE system is to extract useful information from 'raw' data, usually text or webpages. Useful information has many dimensions, but the most important dimension for computational purposes is that the information can be queried, and reasoned about, by machines. One of the reasons why IE was identified as an early problem in communities like natural language processing was because machines are not good at understanding natural languages like English or French due to problems like subtlety, ambiguity and irregularity. Even today, despite rapid advances, machines still cannot read and understand English nearly as well as humans. Thus, an early goal of IE was to extract key pieces of information, such as entities, relations, events and attributes from natural language text. The advent of the Web only made the problem more interesting, since webpages are visually intuitive (if rendered in an compatible browser), but in their raw HTML versions, contain many interesting 'markup' elements like tables, lists, links, images and even dynamic elements like Javascript programs [36]. In Chap. 1, we implicitly demonstrated the results of 'perfect' information extraction when we converted an English sentence (about Fido the dog) to 'equivalent' knowledge graph triples. To take another example, a sentence such as 'Vladimir Putin, President of Russia, attended the G20 meeting', can be represented by the following set of triples: {(Vladimir Putin, presidentOf, Russia), (Russia, isa, Country), (Vladimir Putin, is-a, Person), (G20, is-a, Geopolitical Meeting), (Vladimir Putin, attended, G20)}. Recall that we referred to elements like 'Vladimir Putin' as entities, while elements like 'presidentOf' are generally referred to as relations, relationships or properties. We also introduced the notion of literals (also called attributes or slot fillers depending on both the context and community) in the previous chapter; e.g.,

Lecture Notes in Computer Science, 2012

Entity resolution (ER) is a computationally hard problem of data integration scenarios, where database records have to be grouped according to the real-world entities they belong to. In practice these entities may consist of only a few records from different data sources with typos or historical data. In other cases they may contain significantly more records, especially when we search for entities on a higher level of a concept hierarchy than records. In this paper we give theoretical foundation of a variety of practically important match functions. We show that under these formulations, ER with large entities can be solved efficiently with algorithms based on MapReduce, a distributed computing paradigm. Our algorithm can efficiently incorporate probabilistic and similarity-based record match, enabling flexible match function definition. We demonstrate the usability of our model and algorithm in a real-world insurance ER scenario, where we identify household groups of client records.

Lecture Notes in Computer Science, 2008

This paper presents a method for Entity Disambiguation in Information Extraction from different sources in the web. Once entities and relations between them are extracted, it is needed to determine which ones are referring to the same real-world entity. We model the problem as a graph partitioning problem in order to combine the available information more accurately than a pairwise classifier. Moreover, our method handle uncertain information which turns out to be quite helpful. Two algorithms are trained and compared, one probabilistic and the other deterministic. Both are tuned using genetic algorithms to find the best weights for the set of constraints. Experiments show that graph-based modeling yields better results using uncertain information.

ACM Transactions on Database Systems, 2016

We introduce and develop a declarative framework for entity linking and, in particular, for entity resolution. As in some earlier approaches, our framework is based on a systematic use of constraints. However, the constraints we adopt are link-to-source constraints, unlike in earlier approaches where source-to-link constraints were used to dictate how to generate links. Our approach makes it possible to focus entirely on the intended properties of the outcome of entity linking, thus separating the constraints from any procedure of how to achieve that outcome. The core language consists of link-to-source constraints that specify the desired properties of a link relation in terms of source relations and built-in predicates such as similarity measures. A key feature of the link-to-source constraints is that they employ disjunction, which enables the declarative listing of all the reasons two entities should be linked. We also consider extensions of the core language that capture collec...

Synthesis Lectures on Data Management, 2021

Synthesis Lectures on Data Management is edited by H.V. Jagadish of the University of Michigan. The series publishes 80-150 page publications on topics pertaining to data management. Topics include query languages, database system architectures, transaction management, data warehousing, XML and databases, data stream systems, wide scale data distribution, multimedia data management, data mining, and related subjects.

Security Informatics, 2015

Synthesis lectures on information concepts, retrieval, and services, 2020

This book deals with a hard problem that is inherent to human language: ambiguity. In particular, we focus on author name ambiguity, a type of ambiguity that exists in digital bibliographic repositories, which occurs when an author publishes works under distinct names or distinct authors publish works under similar names. This problem may be caused by a number of reasons, including the lack of standards and common practices, and the decentralized generation of bibliographic content. As a consequence, the quality of the main services of digital bibliographic repositories such as search, browsing, and recommendation may be severely affected by author name ambiguity. The focal point of the book is on automatic methods, since manual solutions do not scale to the size of the current repositories or the speed in which they are updated. Accordingly, we provide an ample view on the problem of automatic disambiguation of author names, summarizing the results of more than a decade of research on this topic conducted by our group, which were reported in more than a dozen publications that received over 900 citations so far, according to Google Scholar. We start by discussing its motivational issues (Chapter 1). Next, we formally define the author name disambiguation task (Chapter 2) and use this formalization to provide a brief, taxonomically organized, overview of the literature on the topic (Chapter 3). We then organize, summarize and integrate the efforts of our own group on developing solutions for the problem that have historically produced state-of-the-art (by the time of their proposals) results in terms of the quality of the disambiguation results. Thus, Chapter 4 covers HHC -Heuristic-based Clustering, an author name disambiguation method that is based on two specific real-world assumptions regarding scientific authorship. Then, Chapter 5 describes SAND -Self-training Author Name Disambiguator and Chapter 6 presents two incremental author name disambiguation methods, namely INDi -Incremental Unsupervised Name Disambiguation and INC-Incremental Nearest Cluster. Finally, Chapter 7 provides an overview of recent author name disambiguation methods that address new specific approaches such as graph-based representations, alternative predefined similarity functions, visualization facilities and approaches based on artificial neural networks. The chapters are followed by three appendices that cover, respectively: (i) a pattern matching function for comparing proper names and used by some of the methods addressed in this book; (ii) a tool for generating synthetic collections of citation records for distinct experimental tasks; and (iii) a number of datasets commonly used to evaluate author name disambiguation methods. In summary, the book organizes a large body of knowledge and work in the area of author name disambiguation in the last decade, hoping to consolidate a solid basis for future developments in the field.

Journal of the Association for Information Science and Technology, 2016

The vast majority of the current author name disambiguation solutions are designed to disambiguate a whole digital library (DL) at once considering the entire repository. However, these solutions besides being very expensive and having scalability problems, also may not benefit from eventual manual corrections, as they may be lost whenever the process of disambiguating the entire repository is required. In the real world, in which repositories are updated on a daily basis, incremental solutions that disambiguate only the newly introduced citation records, are likely to produce improved results in the long run. However, the problem of incremental author name disambiguation has been largely neglected in the literature. In this article we present a new author name disambiguation method, specially designed for the incremental scenario. In our experiments, our new method largely outperforms recent incremental proposals reported in the literature as well as the current state-of-the-art non-incremental method.