AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S.

VERYKIOS 1

Duplicate Record Detection: A Survey

Ahmed K. Elmagarmid

Purdue University

Panagiotis G. Ipeirotis

New York University

Vassilios S. Verykios

University of Thessaly

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 2

Abstract

Often, in the real world, entities have two or more representations in databases. Duplicate records do

not share a common key and/or they contain errors that make duplicate matching a difcult task. Errors

are introduced as the result of transcription errors, incomplete information, lack of standard formats

or any combination of these factors. In this article, we present a thorough analysis of the literature on

duplicate record detection. We cover similarity metrics that are commonly used to detect similar eld

entries, and we present an extensive set of duplicate detection algorithms that can detect approximately

duplicate records in a database. We also cover multiple techniques for improving the efciency and

scalability of approximate duplicate detection algorithms. We conclude with a coverage of existing

tools and with a brief discussion of the big open problems in the area.

Index Terms

duplicate detection, data cleaning, data integration, record linkage, data deduplication, instance

identication, database hardening, name matching, identity uncertainty, entity resolution, fuzzy duplicate

detection, entity matching

I. I NTRODUCTION

Databases play an important role in today's IT based economy. Many industries and systems

depend on the accuracy of databases to carry out operations. Therefore, the quality of the

information (or the lack thereof) stored in the databases, can have signicant cost implications

to a system that relies on information to function and conduct business. In an error-free system

with perfectly clean data, the construction of a comprehensive view of the data consists of

linking –in relational terms, joining– two or more tables on their key elds. Unfortunately, data

often lack a unique, global identier that would permit such an operation. Furthermore, the data

are neither carefully controlled for quality nor dened in a consistent way across different data

sources. Thus, data quality is often compromised by many factors, including data entry errors

(e.g., Microsft instead of Microsoft), missing integrity constraints (e.g., allowing entries such as

EmployeeAge=567), and multiple conventions for recording information (e.g., 44 W. 4th St. vs.

44 West Fourth Street). To make things worse, in independently managed databases not only

the values, but the structure, semantics and underlying assumptions about the data may differ as

well.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 3

Often, while integrating data from different sources to implement a data warehouse, organiza-

tions become aware of potential systematic differences or conicts. Such problems fall under the

umbrella-term data heterogeneity [14]. Data cleaning [77], or data scrubbing [96], refer to the

process of resolving such identication problems in the data. We distinguish between two types

of data heterogeneity: structural and lexical. Structural heterogeneity occurs when the elds of

the tuples in the database are structured differently in different databases. For example, in one

database, the customer address might be recorded in one eld named, say, addr, while in another

database the same information might be stored in multiple elds such as street, city, state, and

zipcode. Lexical heterogeneity occurs when the tuples have identically structured elds across

databases, but the data use different representations to refer to the same real-world object (e.g.,

StreetAddress=44 W. 4th St. vs. StreetAddress=44 West Fourth Street).

In this paper, we focus on the problem of lexical heterogeneity and survey various techniques

which have been developed for addressing this problem. We focus on the case where the input

is a set of structured and properly segmented records, i.e., we focus mainly on cases of database

records. Hence, we do not cover solutions for the various other problems, such that of mirror

detection, in which the goal is to detect similar or identical web pages (e.g., see [13], [18]). Also,

we do not cover solutions for problems such as anaphora resolution [56], in which the problem

is to locate different mentions of the same entity in free text (e.g., that the phrase “President of

the U.S.” refers to the same entity as “George W. Bush”). We should note that the algorithms

developed for mirror detection or for anaphora resolution are often applicable for the task of

duplicate detection. Techniques for mirror detection have been used for detection of duplicate

database records (see, for example, Section V-A.4) and techniques for anaphora resolution are

commonly used as an integral part of deduplication in relations that are extracted from free text

using information extraction systems [52].

The problem that we study has been known for more than ve decades as the record linkage

or the record matching problem [31], [61]–[64], [88] in the statistics community. The goal of

record matching is to identify records in the same or different databases that refer to the same

real-world entity, even if the records are not identical. In slightly ironic fashion, the same problem

has multiple names across research communities. In the database community, the problem

is described as merge-purge [39], data deduplication [78], and instance identication [94];

in the AI community, the same problem is described as database hardening [21] and name

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 4

matching [9]. The names coreference resolution, identity uncertainty, and duplicate detection are

also commonly used to refer to the same task. We will use the term duplicate record detection

in this paper.

The remaining part of this paper is organized as follows: In Section II, we briey discuss

the necessary steps in the data cleaning process, before the duplicate record detection phase.

Then, Section III describes techniques used to match individual elds, and Section IV presents

techniques for matching records that contain multiple elds. Section V describes methods for

improving the efciency of the duplicate record detection process and Section VI presents a few

commercial, off-the-shelf tools used in industry for duplicate record detection and for evaluating

the initial quality of the data and of the matched records. Finally, Section VII concludes the

paper and discusses interesting directions for future research.

II. DATA P REPARATION

Duplicate record detection is the process of identifying different or multiple records that refer

to one unique real-world entity or object. Typically, the process of duplicate detection is preceded

by a data preparation stage, during which data entries are stored in a uniform manner in the

database, resolving (at least partially) the structural heterogeneity problem. The data preparation

stage includes a parsing, a data transformation, and a standardization step. The approaches

that deal with data preparation are also described under the using the term ETL (Extraction,

Transformation, Loading) [43]. These steps improve the quality of the in-ow data and make

the data comparable and more usable. While data preparation is not the focus of this survey, for

completeness we describe briey the tasks performed in that stage. A comprehensive collection

of papers related to various data transformation approaches can be found in [74].

Parsing is the rst critical component in the data preparation stage. Parsing locates, identies

and isolates individual data elements in the source les. Parsing makes it easier to correct,

standardize, and match data because it allows the comparison of individual components, rather

than of long complex strings of data. For example, the appropriate parsing of name and address

components into consistent packets of information is a crucial part in the data cleaning process.

Multiple parsing methods have been proposed recently in the literature (e.g., [1], [11], [53], [71],

[84]) and the area continues to be an active eld of research.

Data transformation refers to simple conversions that can be applied to the data in order for

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 5

them to conform to the data types of their corresponding domains. In other words, this type of

conversion focuses on manipulating one eld at a time, without taking into account the values

in related elds. The most common form of a simple transformation is the conversion of a data

element from one data type to another. Such a data type conversion is usually required when

a legacy or parent application stored data in a data type that makes sense within the context

of the original application, but not in a newly developed or subsequent system. Renaming of

a eld from one name to another is considered data transformation as well. Encoded values in

operational systems and in external data is another problem that is addressed at this stage. These

values should be converted to their decoded equivalents, so records from different sources can

be compared in a uniform manner. Range checking is yet another kind of data transformation

which involves examining data in a eld to ensure that it falls within the expected range, usually

a numeric or date range. Lastly, dependency checking is slightly more involved since it requires

comparing the value in a particular eld to the values in another eld, to ensure a minimal level

of consistency in the data.

Data standardization refers to the process of standardizing the information represented in

certain elds to a specic content format. This is used for information that can be stored in

many different ways in various data sources and must be converted to a uniform representation

before the duplicate detection process starts. Without standardization, many duplicate entries

could erroneously be designated as non-duplicates, based on the fact that common identifying

information cannot be compared. One of the most common standardization applications involves

address information. There is no one standardized way to capture addresses so the same address

can be represented in many different ways. Address standardization locates (using various parsing

techniques) components such as house numbers, street names, post ofce boxes, apartment

numbers and rural routes, which are then recorded in the database using a standardized format

(e.g., 44 West Fourth Street is stored as 44 W4th St.). Date and time formatting and name and

title formatting pose other standardization difculties in a database. Typically, when operational

applications are designed and constructed, there is very little uniform handling of date and time

formats across applications. Because most operational environments have many different formats

for representing dates and times, there is a need to transform dates and times into a standardized

format. Name standardization identies components such as rst names, last names, title and

middle initials and records everything using some standardized convention. Data standardization

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 6

is a rather inexpensive step that can lead to fast identication of duplicates. For example, if

the only difference between two records is the differently recorded address (44 West Fourth

Street vs. 44 W4th St.), then the data standardization step would make the two records identical,

alleviating the need for more expensive approximate matching approaches, that we describe in

the later sections.

After the data preparation phase, the data are typically stored in tables, having comparable

elds. The next step is to identify which elds should be compared. For example, it would not

be meaningful to compare the contents of the eld LastName with the eld Address. Perkowitz

et al. [67] presented a supervised technique for understanding the “semantics” of the elds that

are returned by web databases. The idea was that similar values (e.g. last names) tend to appear

in similar elds. Hence, by observing value overlap across elds, it is possible to parse the

results into elds and discover correspondences across elds at the same time. Dasu et al. [25]

signicantly extend this concept and extract a “signature” from each eld in the database; this

signature summarizes the content of each column in the database. Then, the signatures are used

to identify elds with similar values, elds whose contents are subsets of other elds and so on.

Even after parsing, data standardization, and identication of similar elds, it is not trivial to

match duplicate records. Misspellings and different conventions for recording the same informa-

tion still result in different, multiple representations of a unique object in the database. In the

next section, we describe techniques for measuring the similarity of individual elds, and later,

in Section IV we describe techniques for measuring the similarity of entire records.

III. F IELD M ATCHING T ECHNIQUES

One of the most common sources of mismatches in database entries is the typographical

variations of string data. Therefore, duplicate detection typically relies on string comparison

techniques to deal with typographical variations. Multiple methods have been developed for this

task, and each method works well for particular types of errors. While errors might appear in

numeric elds as well, the related research is still in its infancy.

In this section, we describe techniques that have been applied for matching elds with string

data, in the duplicate record detection context. We also review briey some common approaches

for dealing with errors in numeric data.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 7

A. Character-based similarity metrics

The character-based similarity metrics are designed to handle well typographical errors. In

this section, we cover the following similarity metrics:

• Edit distance,

• Afne gap distance,

• Smith-Waterman distance,

• Jaro distance metric, and

• Q-gram distance.

Edit distance: The edit distance between two strings σ1 and σ2 is the minimum number of

edit operations of single characters needed to transform the string σ1 into σ2 . There are three

types of edit operations:

• Insert a character into the string,

• Delete a character from the string, and

• Replace one character with a different character.

In the simplest form, each edit operation has cost 1. This version of edit distance is also

referred to as Levenshtein distance [49]. The basic dynamic programming algorithm [59] for

computing the edit distance between two strings takes O(|σ1 | · |σ2 |) time for two strings of

length |σ1 | and |σ2 |, respectively. Landau and Vishkin [48] presented an algorithm for detecting

in O(max{|σ1 |, |σ2 |}·k) whether two strings have edit distance less than k . (Notice that if ||σ1 |−
|σ2 || > k then by denition the two strings do not match within distance k , so O(max{|σ1 |, |σ2 |}·
k) ∼ O(|σ1 | · k) ∼ O(|σ2 | · k), for the non-trivial case where ||σ1 | − |σ2 || ≤ k .) Needleman

and Wunsch [60] modied the original edit distance model, and allowed for different costs

for different edit distance operations. (For example, the cost of replacing O with 0 might be

smaller than the cost of replacing f with q.) Ristad and Yiannilos [73] presented a method for

automatically determining such costs from a set of equivalent words that are written in different

ways. The edit distance metrics work well for catching typographical errors, but they are typically

ineffective for other types of mismatches.

Afne gap distance: The edit distance metric described above does not work well when

matching strings that have been truncated, or shortened (e.g., “John R. Smith” vs. “Jonathan

Richard Smith”). The afne gap distance metric [95] offers a solution to this problem by

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 8

Fig. 1. The Pinheiro and Sun similarity metric alignment for the strings σ1 = ABcDeF gH σ2 = AxByDzF .

introducing two extra edit operations: open gap and extend gap. The cost of extending the

gap is usually smaller than the cost of opening a gap, and this results in smaller cost penalties

for gap mismatches than the equivalent cost under the edit distance metric. The algorithm for

computing the afne gap distance requires O(a · |σ1 | · |σ2 |) time, when the maximum length of a

gap a ¿ min{|σ1 |, |σ2 |}. In the general case, the algorithm runs in approximately O(a2 ·|σ1 |·|σ2 |)
steps. Bilenko et al. [9], in a spirit similar to what Ristad and Yiannilos [73] proposed for edit

distance, describe how to train an edit distance model with afne gaps.

Smith-Waterman distance: Smith and Waterman [81] described an extension of edit distance

and afne gap distance, in which mismatches at the beginning and the end of strings have lower

costs than mismatches in the middle. This metric allows for better local alignment of the strings

(i.e., substring matching). Therefore, the strings “Prof. John R. Smith, University of Calgary”

and “John R. Smith, Prof.” can match within short distance using the Smith-Waterman distance,

since the prexes and sufxes are ignored. The distance between two strings can be computed

using a dynamic programming technique, based on the Needleman and Wunsch algorithm [60].

The Smith and Waterman algorithm requires O(|σ1 | · |σ2 |) time and space for two strings of

length |σ1 | and |σ2 |; many improvements have been proposed (e.g., the BLAST algorithm [4]),

mainly in the context of computational biology applications. Pinheiro and Sun [70] proposed a

similar similarity measure, which tries to nd the best character alignment for the two compared

strings σ1 and σ2 , so that the number of character mismatches is minimized. For example, the

strings σ1 = ABcDeF gH and σ2 = AxByDzF H can be aligned as shown in Figure 1.

Jaro distance metric: Jaro [40] introduced a string comparison algorithm that was mainly used

for comparison of last and rst names. The basic algorithm for computing the Jaro metric for

two strings σ1 and σ2 includes the following steps:

1) Compute the string lengths |σ1 | and |σ2 |,

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 9

2) Find the “common characters” c in the two strings; common are all the characters σ1 [j]
and σ2 [j] for which σ1 [i] = σ2 [j] and |i − j| ≤ 12 min{|σ1 |, |σ2 |}.
3) Find the number of transpositions t; the number of transpositions is computed as follows:

We compare the ith common character in σ1 with the ith common character in σ2 . Each

non-matching character is a transposition.

The Jaro comparison value is:

à !
1 c c c − t/2
Jaro(σ1 , σ2 ) = + + . (1)
3 |σ1 | |σ2 | c
From the description of the Jaro algorithm, we can see that the Jaro algorithm requires O(|σ1 |·
|σ2 |) time for two strings of length |σ1 | and |σ2 |, mainly due to the Step 2 that computes the

“common characters” in the two strings. Winkler and Thibaudeau [101] modied the Jaro metric

to give higher weight to prex matches, since prex matches are generally more important for

surname matching.

Q-grams: The q -grams are short character substrings1 of length q of the database strings [89],

[90]. The intuition behind the use of q -grams as a foundation for approximate string matching is

that when two strings σ1 and σ2 are similar they share a large number of q -grams in common.

Given a string σ, its q -grams are obtained by “sliding” a window of length q over the characters

of σ. Since q -grams at the beginning and the end of the string can have fewer than q characters

from σ, the strings are conceptually extended by “padding” the beginning and the end of the

string with q−1 occurrences of a special padding character, not in the original alphabet. With

the appropriate use of hash-based indexes, the average time required for computing the q -gram
overlap between two strings σ1 and σ2 is O(max{|σ1 |, |σ2 |}). Letter q -grams, including trigrams,
bigrams, and/or unigrams, have been used in a variety of ways in text recognition and spelling

correction [47]. One natural extension of q -grams are the positional q -grams [83], which also

record the position of the q -gram in the string. Gravano et al. [34], [35] showed how to use

positional q -grams to locate efciently similar strings within a relational database.

1
The q -grams in our context are dened on the character level. In speech processing and in computational linguistics,

researchers often use the term n-gram, to refer to sequences of n words.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 10

B. Token-based similarity metrics

Character-based similarity metrics work well for typographical errors. However, it is often

the case that typographical conventions lead to rearrangement of words (e.g., “John Smith” vs.

“Smith, John”). In such cases, character-level metrics fail to capture the similarity of the entities.

Token-based metrics try to compensate for this problem.

Atomic strings: Monge and Elkan [57] proposed a basic algorithm for matching text elds

based on atomic strings. An atomic string is a sequence of alphanumeric characters delimited

by punctuation characters. Two atomic strings match if they are equal, or if one is the prex of

the other. Based on this algorithm, the similarity of two elds is the number of their matching

atomic strings divided by their average number of atomic strings.

WHIRL: Cohen [22] described a system named WHIRL that adopts from the information

retrieval the cosine similarity combined with the tf.idf weighting scheme to compute the similarity

of two elds. Cohen separates each string σ into words and each word w is assigned a weight

vσ (w) = log(tf w + 1) · log(idfw )

|D|
where tf w is the number of times that w appears in the eld and idfw is
nw
, where nw is the

number of records in the database D that contain w. The tf.idf weight for a word w in a eld is

high if w appears a large number of times in the eld (large tf w ) and w is a sufciently “rare”

term in the database (large idfw ). For example, for a collection of company names, relatively

infrequent terms such as “AT&T” or “IBM” will have higher idf weights than more frequent

terms such as “Inc.” The cosine similarity of σ1 and σ2 is dened as

P|D|
j=1 vσ1 (j) · vσ2 (j)
sim(σ1 , σ2 ) =
||vσ1 ||2 · ||vσ2 ||2
The cosine similarity metric works well for a large variety of entries, and is insensitive to the

location of words, thus allowing natural word moves and swaps (e.g., “John Smith” is equivalent

to “Smith, John”). Also, introduction of frequent words affects only minimally the similarity of

the two strings due to the low idf weight of the frequent words. For example, “John Smith”

and “ Mr. John Smith” would have similarity close to one. Unfortunately, this similarity metric

does not capture word spelling errors, especially if they are pervasive and affect many of the

words in the strings. For example, the strings “Compter Science Department” and “Deprtment

of Computer Scence” will have zero similarity under this metric. Bilenko et al. [9] suggest the

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 11

SoftTF-IDF metric to solve this problem. In the SoftTF.IDF metric, pairs of tokens that are

2
“similar” (and not necessarily identical) are also considered in the computation of the cosine

similarity. However, the product of the weights for non-identical token pairs is multiplied by the

the similarity of the token pair, which is less than one.

Q-grams with tf.idf: Gravano et al. [36] extended the WHIRL system to handle spelling errors
by using q -grams, instead of words, as tokens. In this setting, a spelling error minimally affects

the set of common q -grams of two strings, so the two strings “Gteway Communications” and

“Comunications Gateway” have high similarity under this metric, despite the block move and

the spelling errors in both words. This metric handles the insertion and deletion of words nicely.

The string “Gateway Communications” matches with high similarity the string “Communications

Gateway International” since the q -grams of the word “International” appear often in the relation
and have low weight.

C. Phonetic similarity metrics

Character-level and token-based similarity metrics focus on the string-based representation of

the database records. However, strings may be phonetically similar even if they are not similar

in a character or token level. For example the word Kageonne is phonetically similar to Cajun

despite the fact that the string representations are very different. The phonetic similarity metrics

are trying to address such issues and match such strings.

Soundex: Soundex, invented by Russell [75], [76], is the most common phonetic coding

scheme. Soundex is based on the assignment of identical code digits to phonetically similar

groups of consonants and is used mainly to match surnames. The rules of Soundex coding are

as follows:

1) Keep the rst letter of the surname as the prex letter and ignore completely all occurrences

of W and H in other positions;

2) Assign the following codes to the remaining letters:

• B, F, P, V → 1
• C, G, J, K, Q, S, X, Z → 2
• D, T → 3

2
The token similarity is measured using a metric that works well for short strings, such as edit distance and Jaro.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 12

• L→4
• M, N → 5
• R→6
3) A, E, I, O, U and Y are not coded but serve as separators (see below);

4) Consolidate sequences of identical codes by keeping only the rst occurrence of the code;

5) Drop the separators;

6) Keep the letter prex and the three rst codes, padding with zeros if there are fewer than

three codes.

Newcombe [61] reports that the Soundex code remains largely unchanged, exposing about

two-thirds of the spelling variations observed in linked pairs of vital records, and that it sets

aside only a small part of the total discriminating power of the full alphabetic surname. The code

is designed primarily for Caucasian surnames, but works well for names of many different origins

(such as those appearing on the records of the U.S. Immigration and Naturalization Service).

However, when the names are of predominantly East Asian origin, this code is less satisfactory,

because much of the discriminating power of these names resides in the vowel sounds, which

the code ignores.

New York State Identication and Intelligence System (NYSIIS): The NYSIIS system, proposed

by Taft [85], differs from Soundex in that it retains information about the position of vowels

in the encoded word by converting most vowels to the letter A. Furthermore, NYSIIS does not

use numbers to replace letters; instead it replaces consonants with other, phonetically similar

letters, thus returning a purely alpha code (no numeric component). Usually the NYSIIS code

for a surname is based on a maximum of nine letters of the full alphabetical name, and the

NYSIIS code itself is then limited to six characters. Tafts [85] compared Soundex with NYSIIS,

using a name database of New York State, and concluded that NYSIIS is 98.72% accurate, while

Soundex is 95.99% accurate for locating surnames. The NYSIIS encoding system is still used

today from the New York State Division of Criminal Justice Services.

Oxford Name Compression Algorithm (ONCA): ONCA [33] is a two-stage technique, designed

to overcome most of the unsatisfactory features of pure Soundex-ing, retaining in parallel the

convenient four-character xed-length format. In the rst step, ONCA uses a British version

of the NYSIIS method of compression. Then, in the second step, the transformed and partially

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 13

compressed name is Soundex-ed in the usual way. This two-stage technique has been used

successfully for grouping similar names together.

Metaphone and Double Metaphone: Philips [68] suggested the Metaphone algorithm as a

better alternative to Soundex. Philips suggested using 16 consonant sounds that can describe a

large number of sounds used in many English and non-English words. Double Metaphone [69] is

a better version of Metaphone, improving some encoding choices made in the initial Metaphone

and allowing multiple encodings for names that have various possible pronunciations. For such

cases, all possible encodings are tested when trying to retrieve similar names. The introduction

of multiple phonetic encodings greatly enhances the matching performance, with rather small

overhead. Philips suggested that, at most, 10% of American surnames have multiple encodings.

D. Numeric Similarity Metrics

While multiple methods exist for detecting similarities of string-based data, the methods for

capturing similarities in numeric data are rather primitive. Typically, the numbers are treated

as strings (and compared using the metrics described above) or simple range queries, which

locate numbers with similar values. Koudas et al. [46] suggest, as direction for future research,

consideration of the distribution and type of the numeric data, or extending the notion of cosine

similarity for numeric data [2] to work well for duplicate detection purposes.

E. Concluding Remarks

The large number of eld comparison metrics reects the large number of errors or transfor-

mations that may occur in real-life data. Unfortunately, there are very few studies that compare

the effectiveness of the various distance metrics presented here. Yancey [103] shows that the

Jaro-Winkler metric works well for name matching tasks for data coming from U.S. census. A

notable comparison effort is the work of Bilenko et al. [9], who compare the effectiveness of

character-based and token-based similarity metrics. They show that the Monge-Elkan metric has

the highest average performance across data sets and across character-based distance metrics.

They also show that the SoftTF.IDF metric works better than any other metric. However, Bilenko

et al. emphasize that no single metric is suitable for all data sets. Even metrics that demonstrate

robust and high performance for some data sets can perform poorly on others. Hence, they

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 14

advocate more exible metrics that can accommodate multiple similarity comparisons (e.g., [9],

[87]). In the next section we review such approaches.

IV. D ETECTING D UPLICATE R ECORDS

In the previous section we described methods that can be used to match individual elds of

a record. In most real-life situations, however, the records consist of multiple elds, making the

duplicate detection problem much more complicated. In this section, we review methods that are

used for matching records with multiple elds. The presented methods can be broadly divided

into two categories:

• Approaches that rely on training data to “learn” how to match the records. This category

includes (some) probabilistic approaches and supervised machine learning techniques.

• Approaches that rely on domain knowledge or on generic distance metrics to match records.

This category includes approaches that use declarative languages for matching, and ap-

proaches that devise distance metrics appropriate for the duplicate detection task.

The rest of this section is organized as follows: initially, in Section IV-A we describe the

notation. In Section IV-B we present probabilistic approaches for solving the duplicate detection

problem. In Section IV-C we list approaches that use supervised machine learning techniques and

in Section IV-D we describe variations based on active learning methods. Section IV-E describes

distance-based methods and Section IV-F describes declarative techniques for duplicate detec-

tion. Finally, Section IV-G covers unsupervised machine learning techniques, and Section IV-H

provides some concluding remarks.

A. Notation

We use A and B to denote the tables that we want to match, and we assume, without loss

of generality, that A and B have n comparable elds. In the duplicate detection problem, each

tuple pair hα, βi, (α ∈ A, β ∈ B), is assigned to one of the two classes M and U. The class M
contains the record pairs that represent the same entity (“match”) and the class U contains the

record pairs that represent two different entities (“non-match”).

We represent each tuple pair hα, βi as a random vector x = [x1 , . . . , xn ]T with n components

that correspond to the n comparable elds of A and B. Each xi shows the level of agreement

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 15

of the ith eld for the records α and β. Many approaches use binary values for the xi 's and set

xi = 1 if eld i agrees and let xi = 0 if eld i disagrees.

B. Probabilistic Matching Models

Newcombe et al. [64] were the rst to recognize duplicate detection as a Bayesian inference

problem. Then, Fellegi and Sunter [31] formalized the intuition of Newcombe et al. and intro-

duced the notation that we use, which is also commonly used in duplicate detection literature.

The comparison vector x is the input to a decision rule that assigns x to U or to M. The

main assumption is that x is a random vector whose density function is different for each of

the two classes. Then, if the density function for each class is known, the duplicate detection

problem becomes a Bayesian inference problem. In the following sections, we will discuss

various techniques that have been developed for addressing this (general) decision problem.

1) The Bayes Decision Rule for Minimum Error: Let x be a comparison vector, randomly

drawn from the comparison space that corresponds to the record pair hα, βi. The goal is to

determine whether hα, βi ∈ M or hα, βi ∈ U . A decision rule, based simply on probabilities,

can be written as follows:



 M if p(M |x) ≥ p(U |x)
hα, βi ∈ (2)

 U otherwise

This decision rule indicates that if the probability of the match class M , given the comparison
vector x, is larger than the probability of the non-match class U, then x is classied to M, and

vice versa. By using the Bayes theorem, the previous decision rule may be expressed as:


 M p(x|M ) p(U )
if l(x) = p(x|U
) ≥ p(M )
hα, βi ∈ (3)

 U otherwise

The ratio
p(x|M )
l(x) = (4)
p(x|U )
p(U )
is called the likelihood ratio. The ratio denotes the threshold value of the likelihood ratio
p(M )

for the decision. We refer to the decision rule in Equation 3 as the Bayes test for minimum

error. It can be easily shown [38] that the Bayes test results in the smallest probability of error,

and it is in that respect an optimal classier. Of course this holds only when the distributions

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 16

of p(x|M ), p(x|U ) and the priors p(U ) and p(M ) are known; this unfortunately is very rarely

the case.

One common approach, usually called Naive Bayes, to compute the distributions of p(x|M )
and p(x|U ) is to make a conditional independence assumption, and postulate that the probabilities
p(xi |M ) and p(xj |M ) are independent if i 6= j . (Similarly, for p(xi |U ) and p(xj |U ).) In that

case, we have
n
Y
p(x|M ) = p(xi |M )
i=i
Yn
p(x|U ) = p(xi |U )
i=i

The values of p(xi |M ) and p(xi |U ) can be computed using a training set of pre-labeled record

pairs. However, the probabilistic model can also be used without using training data. Jaro [41]

used a binary model for the values of xi (i.e., if the eld i “matches” xi = 1 , else xi = 0 ) and

suggested using an expectation maximization (EM) algorithm [26] to compute the probabilities

p(xi = 1|M ). The probabilities p(xi = 1|U ) can be estimated by taking random pairs of records

(which are with high probability in U ).

When the conditional independence is not a reasonable assumption, then Winkler [97] sug-

gested using the general expectation maximization algorithm to estimate p(x|M ), p(x|U ). In [99],
Winkler claims that the general, unsupervised EM algorithm works well under ve conditions:

1) the data contain a relatively large percentage of matches (more than 5%),

2) the matching pairs are “well-separated” from the other classes,

3) the rate of typographical errors is low,

4) there are sufciently many redundant identiers to overcome errors in other elds of the

record, and

5) the estimates computed under the conditional independence assumption result in good

classication performance.

Winkler [99] shows how to relax the assumptions above (including the conditional indepen-

dence assumption) and still get good matching results. Winkler shows that a semi-supervised

model, which combines labeled and unlabeled data (similar to Nigam et al. [65]), performs better

than purely unsupervised approaches. When no training data is available, unsupervised EM works

well, even when a limited number of interactions is allowed between the variables. Interestingly,

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 17

the results under the independence assumption are not considerably worse compared to the case

in which the EM model allows variable interactions.

Du Bois [28] pointed out the importance of the fact that many times elds have missing

(null) values and proposed a different method to correct mismatches that occur due to missing

values. Du Bois suggested using a new comparison vector x∗ with dimension 2n instead of the

n-dimensional comparison vector x, such that

x∗ = (x1 , x2 , . . . , xn , x1 y1 , x2 y2 , . . . , xn yn ) (5)

where 

 1 if the i-th eld on both records is present,
yi =  (6)
 0 otherwise.

Using this representation, mismatches that occur due to missing data are typically discounted,

resulting in improved duplicate detection performance. Du Bois proposed using an independence

model to learn the distributions of p(xi yi |M ) and p(xi yi |U ) by using a set of pre-labeled training
record pairs.

2) The Bayes Decision Rule for Minimum Cost: Often, in practice, the minimization of the

probability of error is not the best criterion for creating decision rules, as the misclassications

of M and U samples may have different consequences. Therefore, it is appropriate to assign a

cost cij to each situation, which is the cost of deciding that x belongs to the class i when x
actually belongs to the class j. Then, the expected costs rM (x) and rU (x) of deciding that x
belongs to the class M and U, respectively, are:

rM (x) = cM M · p(M |x) + cM U · p(U |x)

rU (x) = cU M · p(M |x) + cU U · p(U |x)

In that case, the decision rule for assigning x to M becomes:



 M if rM (x) < rU (x)
hα, βi ∈ (7)

 U otherwise

It can be easily proved [29] that the minimum cost decision rule for the problem can be stated

as: 

 M (cM U −cU U )·p(U )
if l(x) > (cU M −cM M )·p(M )
hα, βi ∈ (8)

 U otherwise

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 18

Comparing the minimum error and minimum cost decision rule, we notice that the two decision

rules become the same for the special setting of the cost functions to cU M − cM M = cM U − cU U .
In this case, the cost functions are termed symmetrical. For a symmetrical cost function, the cost

becomes the probability of error and the Bayes test for minimum cost specically addresses and

minimizes this error.

3) Decision with a Reject Region: Using the Bayes Decision rule when the distribution

parameters are known leads to optimal results. However, even in an ideal scenario, when the

likelihood ratio l(x) is close to the threshold, the error (or cost) of any decision is high [29].

Based on this well-known and general idea in decision theory, Fellegi and Sunter [31], suggested

adding an extra “reject” class in addition to the classes M and U. The reject class contained

record pairs for which it is not possible to make any denite inference, and a “clerical review”

is necessary. These pairs are examined manually by experts to decide whether they are true

matches or not. By setting thresholds for the conditional error on M and U, we can dene the

reject region and the reject probability, which measure the probability of directing a record pair

to an expert for review.

Tepping [88] was the rst to suggest a solution methodology focusing on the costs of the

decision. He presented a graphical approach for estimating the likelihood thresholds. Verykios

et al. [93] developed a formal framework for the cost-based approach taken by Tepping which

shows how to compute the thresholds for the three decision areas when the costs and the priors

P (M ) and P (U ) are known.

The “reject region” approach can be easily extended to a larger number of decision areas [92].

The main problem with such a generalization is appropriately ordering the thresholds which

determine the regions in a way that no region disappears.

C. Supervised and Semi-Supervised Learning

The probabilistic model uses a Bayesian approach to classify record pairs into two classes,

M and U. This model was widely used for duplicate detection tasks, usually as an application

of the Fellegi-Sunter model. While the Fellegi-Sunter approach dominated the eld for more

than two decades, the development of new classication techniques in the machine learning

and statistics communities prompted the development of new deduplication techniques. The

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 19

supervised learning systems rely on the existence of training data in the form of record pairs,

pre-labeled as matching or not.

One set of supervised learning techniques treat each record pair hα, βi independently, similarly
to the probabilistic techniques of Section IV-B. Cochinwala et al. [19] used the well-known

CART algorithm [12], which generates classication and regression trees, a linear discriminant

algorithm [38], which generates linear combination of the parameters for separating the data

according to their classes, and a “vector quantization” approach, which is a generalization of

nearest neighbor algorithms. The experiments which were conducted indicate that CART has

the smallest error percentage. Bilenko et al. [9] use SVMlight [42] to learn how to merge the

matching results for the individual elds of the records. Bilenko et al. showed that the SVM

approach usually outperforms simpler approaches, such as treating the whole record as one large

eld. A typical post-processing step for these techniques (including the probabilistic techniques

of Section IV-B) is to construct a graph for all the records in the database, linking together the

matching records. Then, using the transitivity assumption, all the records that belong to the same

connected component are considered identical [58].

The transitivity assumption can sometimes result in inconsistent decisions. For example, hα, βi
and hα, γi can be considered matches, but hβ, γi not. Partitioning such “inconsistent” graphs with
the goal of minimizing inconsistencies is an NP-complete problem [6]. Bansal et al. [6] propose

a polynomial approximation algorithm that can partition such a graph, identifying automatically

the clusters and the number of clusters in the dataset. Cohen and Richman [23] proposed a

supervised approach in which the system learns from training data how to cluster together

records that refer to the same real-world entry. The main contribution of this approach is the

adaptive distance function which is learned from a given set of training examples. McCallum

and Wellner [55] learn the clustering method using training data; their technique is equivalent

to a graph partitioning technique that tries to nd the min-cut and the appropriate number of

clusters for the given data set, similarly to the work of Bansal et al. [6].

The supervised clustering techniques described above have records as nodes for the graph.

Singla and Domingos [80] observed that by using attribute values as nodes, it is possible

to propagate information across nodes and improve duplicate record detection. For example,

if the records hGoogle, MountainView , CAi and hGoogleInc., MountainView , Californiai are

deemed equal, then CA and California are also equal, and this information can be useful

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 20

for other record comparisons. The underlying assumption is that the only differences are due

to different representations of the same entity (e.g., “Google” and “Google Inc.”) and that

there is no erroneous information in the attribute values (e.g., by mistake someone entering

Bismarck , ND as the location of Google headquarters). Pasula et al. [66] propose a semi-

supervised probabilistic relational model that can handle a generic set of transformations. While

the model can handle a large number of duplicate detection problems, the use of exact inference

results in a computationally intractable model. Pasula et al. propose to use a Markov Chain

Monte Carlo (MCMC) sampling algorithm to avoid the intractability issue. However, it is unclear

whether techniques that rely on graph-based probabilistic inference can scale well for data sets

with hundreds of thousands of records.

D. Active-Learning-Based Techniques

One of the problems with the supervised learning techniques is the requirement for a large

number of training examples. While it is easy to create a large number of training pairs that are

either clearly non-duplicates or clearly duplicates, it is very difcult to generate ambiguous cases

that would help create a highly accurate classier. Based on this observation, some duplicate

detection systems used active learning techniques [24] to automatically locate such ambiguous

pairs. Unlike an “ordinary” learner that is trained using a static training set, an “active” learner

actively picks subsets of instances from unlabeled data, which, when labeled, will provide the

highest information gain to the learner.

Sarawagi and Bhamidipaty [78] designed ALIAS, a learning based duplicate detection system,

that uses the idea of a “reject region” (see Section IV-B.3) to signicantly reduce the size of

the training set. The main idea behind ALIAS is that most duplicate and non-duplicate pairs

are clearly distinct. For such pairs, the system can automatically categorize them in U and M
without the need of manual labeling. ALIAS requires humans to label pairs only for cases where

the uncertainty is high. This is similar to the “reject region” in the Fellegi and Sunter model,

which marked ambiguous cases as cases for clerical review.

ALIAS starts with small subsets of pairs of records designed for training, which have been

characterized as either matched or unique. This initial set of labeled data forms the training data

for a preliminary classier. In the sequel, the initial classier is used for predicting the status of

unlabeled pairs of records. The initial classier will make clear determinations on some unlabeled

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 21

instances but lack determination on most. The goal is to seek out from the unlabeled data pool

those instances which, when labeled, will improve the accuracy of the classier at the fastest

possible rate. Pairs whose status is difcult to determine serve to strengthen the integrity of the

learner. Conversely, instances in which the learner can easily predict the status of the pairs do not

have much effect on the learner. Using this technique, ALIAS can quickly learn the peculiarities

of a data set and rapidly detect duplicates using only a small number of training data.

Tejada et al. [86], [87] used a similar strategy and employed decision trees to teach rules

for matching records with multiple elds. Their method suggested that by creating multiple

classiers, trained using slightly different data or parameters, it is possible to detect ambiguous

cases and then ask the user for feedback. The key innovation in this work is the creation of

several redundant functions and the concurrent exploitation of their conicting actions in order

to discover new kinds of inconsistencies among duplicates in the data set.

E. Distance-Based Techniques

Even active learning techniques require some training data or some human effort to create the

matching models. In the absence of such training data or ability to get human input, supervised

and active learning techniques are not appropriate. One way of avoiding the need for training

data is to dene a distance metric for records, which does not need tuning through training data.

Using the distance metric and an appropriate matching threshold, it is possible to match similar

records, without the need for training.

One approach is to treat a record as a long eld, and use one of the distance metrics described

in Section III to determine which records are similar. Monge and Elkan [57], [58] proposed a

string matching algorithm for detecting highly similar database records. The basic idea was to

apply a general purpose eld matching algorithm, especially one that is able to account for gaps in

the strings, to play the role of the duplicate detection algorithm. Similarly, Cohen [20] suggested

to use the tf.idf weighting scheme (see Section III-B), together with the cosine similarity metric

to measure the similarity of records. Koudas et al. [46] presented some practical solutions to

problems encountered during the deployment of such a string-based duplicate detection system

at AT&T.

Distance-based approaches that conate each record in one big eld may ignore important

information that can be used for duplicate detection. A simple approach is to measure the distance

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 22

between individual elds, using the appropriate distance metric for each eld, and then compute

the weighted distance [27] between the records. In this case, the problem is the computation

of the weights, and the overall setting becomes very similar to the probabilistic setting that we

discussed in Section IV-B. An alternative approach, proposed by Guha et al. [37] is to create a

distance metric that is based on ranked list merging. The basic idea is that if we compare only

one eld from the record, the matching algorithm can easily nd the best matches and rank

them according to their similarity, putting the best matches rst. By applying the same principle

for all the elds, we can get, for each record, n ranked lists of records, one for each eld.

Then, the goal is to create a rank of records that has the minimum aggregate rank distance when

compared to all the n lists. Guha et al. map the problem into the minimum cost perfect matching
problem, and develop then efcient solutions for identifying the top-k matching records. The rst

solution is based on the Hungarian Algorithm [3], a graph-theoretic algorithm that solves the

minimum cost perfect matching problem. Guha et al. also present the Successive Shortest Paths

algorithm that works well for smaller values of k and is based on the idea that it is not required

to examine all potential matches to identify the top-k matches. Both of the proposed algorithms

are implemented in T-SQL and are directly deployable over existing relational databases.

The distance-based techniques described so far, treat each record as a at entity, ignoring

the fact that data is often stored in relational databases, in multiple tables. Ananthakrishna

et al. [5] describe a similarity metric that uses not only the textual similarity, but the “co-

occurrence” similarity of two entries in a database. For example, the entries in the state column

“CA” and “California” have small textual similarity; however, the city entries “San Francisco,”

“Los Angeles,” “San Diego” and so on, often have foreign keys that point both to “CA”

and “California.” Therefore, it is possible to infer that “CA” and “California” are equivalent.

Ananthakrishna et al. show that by using “foreign key co-occurrence” information, they can

substantially improve the quality of duplicate detection in databases that use multiple tables to

store the entries of a record. This approach is conceptually similar to the work of Perkowitz et

al. [67] and of Dasu et al. [25], which examine the contents of elds to locate the matching

elds across two tables (see Section II).

Finally, one of the problems of the distance-based techniques is the need to dene the

appropriate value for the matching threshold. In the presence of training data, it is possible to nd

the appropriate threshold value. However, this would nullify the major advantage of distance-

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 23

based techniques, which is the ability to operate without training data. Recently, Chaudhuri et

al. [16] proposed a new framework for distance-based duplicate detection, observing that the

distance thresholds for detecting real duplicate entries is different from each database tuple. To

detect the appropriate threshold, Chaudhuri et al. observed that entries that correspond to the

same real-world object but have different representation in the database, tend to (1) have small

distances from each other (compact set property), and to (2) have only a small number of other

neighbors within a small distance (sparse neighborhood property). Furthermore, Chaudhuri et

al. propose an efcient algorithm for computing the required threshold for each object in the

database, and show that the quality of the results outperforms approaches that rely on a single,

global threshold.

F. Rule-based Approaches

A special case of distance-based approaches is the use of rules to dene whether two records

are the same or not. Rule-based approaches can be considered as distance-based techniques,

where the distance of two records is either 0 or 1. Wang and Madnick [94] proposed a rule-

based approach for the duplicate detection problem. For cases in which there is no global key,

Wang and Madnick suggest the use of rules developed by experts to derive a set of attributes

that collectively serve as a “key” for each record. For example, an expert might dene rules

such as

IF age< 22 THEN status = undergraduate

ELSE status = graduate
IF distanceFromHome > 10 THEN transportation = car
ELSE transportation = bicycle
By using such rules, Wang and Madnick hoped to generate unique keys that can cluster

multiple records that represent the same real-world entity. Lim et al. [50] also used a rule-

based approach, but with the extra restriction that the result of the rules must always be correct.

Therefore, the rules should not be heuristically-dened but should reect absolute truths and

serve as functional dependencies.

Hernández and Stolfo [39] further developed this idea and derived an equational theory that

dictates the logic of domain equivalence. This equational theory species an inference about the

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 24

similarity of the records. For example, if two persons have similar name spellings, and these

persons have the same address, we may infer that they are the same person. Specifying such an

inference in the equational theory requires declarative rule language. For example, the following

is a rule that exemplies one axiom of the equational theory developed for an employee database:

FORALL (r1,r2) in EMPLOYEE

IF r1.name is similar to r2.name AND
r1.address = r2.address
THEN r1 matches r2
Note that “similar to” is measured by one of the string comparison techniques (Section III),

and “matches” means to declare that those two records are matched and therefore represent the

same person.

AJAX [32] is a prototype system that provides a declarative language for specifying data

cleaning programs, consisting of SQL statements enhanced with a set of primitive operations

to express various cleaning transformations. AJAX provides a framework wherein the logic of

a data cleaning program is modeled as a directed graph of data transformations starting from

some input source data. Four types of data transformations are provided to the user of the

system. The mapping transformation standardizes data, the matching transformation nds pairs

of records that probably refer to the same real object, the clustering transformation groups

together matching pairs with a high similarity value, and nally, the merging transformation

collapses each individual cluster into a tuple of the resulting data source.

It is noteworthy that such rule-based approaches, which require a human expert to devise

meticulously crafted matching rules, typically result in systems with high accuracy. However,

the required tuning requires extremely high manual effort from the human experts, and this effort

makes the deployment of such systems difcult in practice. Currently, the typical approach is to

use a system that generates matching rules from training data (see Sections IV-C and IV-D) and

then manually tune the automatically generated rules.

G. Unsupervised Learning

As we mentioned earlier, the comparison space consists of comparison vectors which contain

information about the differences between elds in a pair of records. Unless some information

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 25

exists about which comparison vectors correspond to which category (match, non-match, or

possible-match), the labeling of the comparison vectors in the training data set should be done

manually. One way to avoid manual labeling of the comparison vectors is to use clustering

algorithms, and group together similar comparison vectors. The idea behind most unsupervised

learning approaches for duplicate detection is that similar comparison vectors correspond to the

same class.

The idea of unsupervised learning for duplicate detection has its roots in the probabilistic model

proposed by Fellegi and Sunter (see Section IV-B). As we discussed in Section IV-B, when there

are no training data to compute the probability estimates, it is possible to use variations of the

Expectation Maximization algorithm to identify appropriate clusters in the data.

Verykios et al. [91] propose the use of a bootstrapping technique based on clustering to learn

matching models. The basic idea, also known as co-training [10], is to use very few labeled

data, and then use unsupervised learning techniques to label appropriately the data with unknown

labels. Initially, Verykios et al. treat each entry of the comparison vector (which corresponds to

the result of a eld comparison) as a continuous, real variable. Then, they partition the comparison

space into clusters by using the AutoClass [17] clustering tool. The basic premise is that each

cluster contains comparison vectors with similar characteristics. Therefore all the record pairs

in the cluster belong to the same class (matches, non-matches, or possible-matches). Thus, by

knowing the real class of only a few vectors in each cluster, it is possible to infer the class of

all vectors in the cluster, and therefore mark the corresponding record pairs as matches or not.

Elfeky et al. [30] implemented this idea in TAILOR, a toolbox for detecting duplicate entries

in data sets. Verykios et al. show that the classiers generated using the new, larger training set

have high accuracy, and require only a minimal number of pre-labeled record pairs.

Ravikumar and Cohen [72] follow a similar approach and propose a hierarchical, graphical

model for learning to match record pairs. The foundation of this approach is to model each

eld of the comparison vector as a latent binary variable which shows whether the two elds

match or not. The latent variable then denes two probability distributions for the values of the

corresponding “observed” comparison variable. Ravikumar and Cohen show that it is easier to

learn the parameters of a hierarchical model than to attempt to directly model the distributions

of the real-valued comparison vectors. Bhattacharya and Getoor [8] propose to use the Latent

Dirichlet Allocation generative model to perform duplicate detection. In this model, the latent

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 26

variable is a unique identier for each entity in the database.

H. Concluding Remarks

There are multiple techniques for duplicate record detection. We can divide the techniques into

two broad categories: ad-hoc techniques that work quickly on existing relational databases, and

more “principled” techniques that are based on probabilistic inference models. While probabilistic

methods outperform ad-hoc techniques in terms of accuracy, the ad-hoc techniques work much

faster and can scale to databases with hundreds of thousands of records. Probabilistic inference

techniques are practical today only for data sets that are one or two orders of magnitude smaller

than the data sets handled by ad-hoc techniques. A promising direction for future research is

to devise techniques that can substantially improve the efciency of approaches that rely on

machine learning and probabilistic inference.

A question that is unlikely to be resolved soon is the question of which of the presented

methods should be used for a given duplicate detection task. Unfortunately, there is no clear

answer to this question. The duplicate record detection task is highly data-dependent and it is

unclear if we will ever see a technique dominating all others across all data sets. The problem

of choosing the best method for duplicate data detection is very similar to the problem of model

selection and performance prediction for data mining: we expect that progress in that front will

also benet the task of selecting the best method for duplicate detection.

V. I MPROVING THE E FFICIENCY OF D UPLICATE D ETECTION

So far, in our discussion of methods for detecting whether two records refer to the same real-

world object, we have focused mainly on the quality of the comparison techniques and not on

the efciency of the duplicate detection process. Now, we turn to the central issue of improving

the speed of duplicate detection.

An elementary technique for discovering matching entries in tables A and B is to execute a

“nested-loop” comparison, i.e., to compare every record of table A with every record in table B .
Unfortunately, such strategy requires a total of |A| · |B| comparisons, a cost that is prohibitively

expensive even for the moderately-sized tables. In Section V-A we describe techniques that

substantially reduce the number of required comparisons.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 27

Another factor that can lead to increased computation expense is the cost required for a single

comparison. It is not uncommon for a record to contain tens of elds. Therefore, each record

comparison requires multiple eld comparisons and each eld comparison can be expensive. For

example, computing the edit distance between two long strings σ1 and σ2 , respectively, has a

cost of O(|σ1 | · |σ2 |); just checking if they are within a prespecied edit distance threshold k
can reduce the complexity to O(max{|σ1 |, |σ2 |} · k) (see Section III-A). We examine some of

the methods that can be used to reduce the cost of record comparison in Section V-B.

A. Reducing the Number of Record Comparisons

1) Blocking: One “traditional” method for identifying identical records in a database table is

to scan the table and compute the value of a hash function for each record. The value of the

hash function denes the “bucket” to which this record is assigned. By denition, two records

that are identical will be assigned to the same bucket. Therefore, in order to locate duplicates, it

is enough to compare only the records that fall into the same bucket for matches. The hashing

technique cannot be used directly for approximate duplicates, since there is no guarantee that the

hash value of two similar records will be the same. However, there is an interesting counterpart

of this method, named blocking.

As discussed above with relation to utilizing the hash function, blocking typically refers to

the procedure of subdividing les into a set of mutually exclusive subsets (blocks) under the

assumption that no matches occur across different blocks. A common approach to achieving

these blocks is to use a function such as Soundex, NYSIIS, or Metaphone (see Section III-C)

on highly discriminating elds (e.g., last name) and compare only records that have similar, but

not necessarily identical, elds.

Although blocking can increase substantially the speed of the comparison process, it can also

lead to an increased number of false mismatches due to the failure of comparing records that

do not agree on the blocking eld. It can also lead to an increased number of missed matches

due to errors in the blocking step that placed entries in the wrong buckets, thereby preventing

them from being compared to actual matching entries. One alternative is to execute the duplicate

detection algorithm in multiple runs, each time using a different eld for blocking. This approach

can reduce substantially the probability of false mismatches, with a relatively small increase in

the running time.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 28

2) Sorted Neighborhood Approach: Hernández and Stolfo [39] describe the so-called sorted

neighborhood approach. The method consists of the following three steps:

• Create key: A key for each record in the list is computed by extracting relevant elds or

portions of elds.

• Sort data: The records in the database are sorted by using the key found in the rst step.

A sorting key is dened to be a sequence of attributes, or a sequence of sub-strings within

the attributes, chosen from the record in an ad hoc manner. Attributes that appear rst in

the key have a higher priority than those that appear subsequently.

• Merge: A xed size window is moved through the sequential list of records in order to

limit the comparisons for matching records to those records in the window. If the size of

the window is w records then every new record that enters that window is compared with

the previous w − 1 records to nd “matching” records. The rst record in the window slides
out of it.

The sorted neighborhood approach relies on the assumption that duplicate records will be close

in the sorted list, and therefore will be compared during the merge step. The effectiveness of

the sorted neighborhood approach is highly dependent upon the comparison key that is selected

to sort the records. In general, no single key will be sufcient to sort the records in such a way

that all the matching records can be detected. If the error in a record occurs in the particular

eld or portion of the eld that is the most important part of the sorting key, there is a very

small possibility that the record will end up close to a matching record after sorting.

To increase the number of similar records merged, Hernández and Stolfo implemented a

strategy for executing several independent runs of the sorted-neighborhood method (presented

above) by using a different sorting key and a relatively small window each time. This strategy

is called the multi-pass approach. This method is similar in spirit to the multiple-run blocking

approach described above. Each independent run produces a set of pairs of records that can be

merged. The nal results, including the transitive closure of the records matched in different

passes, is subsequently computed.

3) Clustering and Canopies: Monge and Elkan [58] try to improve the performance of a

basic “nested-loop” record comparison, by assuming that duplicate detection is transitive. This

means that if α is deemed duplicate of β and β is deemed duplicate of γ then α and γ are also

duplicates. Under the assumption of transitivity, the problem of matching records in a database

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 29

can be described in terms of determining the connected components of an undirected graph.

At any time, the connected components of the graph correspond to the transitive closure of

the “record matches” relationships discovered so far. Monge and Elkan [58] use a union-nd

structure to efciently compute the connected components of the graph. During the Union step,

duplicate records are “merged” into a cluster and only a “representative” of the cluster is kept for

subsequent comparisons. This reduces the total number of record comparisons, without reducing

substantially the accuracy of the duplicate detection process. The concept behind this approach,

is that if a record α is not similar to a record β already in the cluster, then it will not match the

other members of the cluster either.

McCallum et al. [54] propose the use of canopies for speeding up the duplicate detection

process. The basic idea is to use a cheap comparison metric to group records into overlapping

clusters called canopies. (This is in contrast to blocking that requires hard, non-overlapping

partitions.) After the rst step, the records are then compared pairwise, using a more expensive

similarity metric that leads to better qualitative results. The assumption behind this method is that

there is an inexpensive similarity function that can be used as a “quick-and-dirty” approximation

for another, more expensive function. For example, if two strings have length difference larger

than 3, then their edit distance cannot be smaller than 3. In that case, the length comparison serves

as a cheap (canopy) function for the more expensive edit distance. Cohen and Richman [23]

propose the tf.idf similarity metric as a canopy distance, and then use multiple (expensive)

similarity metrics to infer whether two records are duplicates. Gravano et al. [35] propose using

the string lengths and the number of common q -grams of two strings as canopies (lters according
to [35]) for the edit distance metric, which is expensive to compute in a relational database.

The advantage of this technique is that the canopy functions can be evaluated efciently using

vanilla SQL statements. In a similar fashion, Chaudhuri et al. [15] propose using an indexable

canopy function for easily identifying similar tuples in a database. Baxter et al. [7] perform

an experimental comparison of canopy-based approaches with traditional blocking and show

that the exible nature of canopies can signicantly improve the quality and speed of duplicate

detection.

4) Set Joins: Another direction towards efciently implementing data cleaning operations is

to speed-up the execution of set operations: large number of similarity metrics, discussed in

Section III, use set operations as part of the overall computation. Running set operations on all

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 30

pair combinations is a computationally expensive operation and is typically unnecessary. For

data cleaning applications, the interesting pairs are only those in which the similarity value is

high. Many techniques use this property and suggest algorithms for fast computation of set-based

operations on a set of records.

Cohen [20] proposed using a set of in-memory inverted indexes together with an A∗ search

algorithm to locate the top-k most similar pairs, according to the cosine similarity metric. Soffer

et al. [82], mainly in the context of information retrieval, suggest pruning the inverted index,

removing terms with low weights since they do not contribute much to the computation of

the tf.idf cosine similarity. Gravano et al. [36] present an SQL-based approach that is anal-

ogous to the approach of Soffer et al. [82], and allows fast computation of cosine similarity

within an RDBMS. Mamoulis [51] presents techniques for efciently processing a set join in

a database, focusing on the containment and non-zero-overlap operators. Mamoulis shows that

inverted indexes are typically superior to approaches based on signature les, conrming earlier

comparison studies [104]. Sarawagi and Kirpal [79] extend the set joins approach to a large

number of similarity predicates that use set joins. The Probe-Cluster approach of Sarawagi

and Kirpal works well in environments with limited main memory, and can be used to compute

efciently a large number of similarity predicates, in contrast to previous approaches which were

tuned for a smaller number of similarity predicates (e.g., set containment, or cosine similarity).

Furthermore, Probe-Cluster returns exact values for the similarity metrics, in contrast to previous

approaches which used approximation techniques.

B. Improving the Efciency of Record Comparison

So far, we have examined techniques that reduce the number of required record comparisons

without compromising the quality of the duplicate detection process. Another way of improving

the efciency of duplicate detection is to improve the efciency of a single record comparison.

Next, we review some of these techniques.

When comparing two records, after having computed the differences of only a small portion

of the elds of two records, it may be obvious that the pair does match, irrespective of the results

of further comparison. Therefore, it is paramount to determine the eld comparison for a pair

of records as soon as possible to avoid wasting additional, valuable time. The eld comparisons

should be terminated when even complete agreement of all the remaining elds cannot reverse

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 31

the unfavorable evidence for the matching of the records [62]. To make the early termination

work, the global likelihood ratio for the full agreement of each of the identiers should be

calculated. At any given point in the comparison sequence, the maximum collective favorable

evidence, which could be accumulated from that point forward, will indicate what improvement

in the overall likelihood ratio might conceivably result if the comparisons were continued.

Verykios et al. [91] propose a set of techniques for reducing the complexity of record compari-

son. The rst step is to apply a feature subset selection algorithm for reducing the dimensionality

of the input set. By using a feature selection algorithm (e.g., [44]) as a preprocessing step the

record comparison process uses only a small subset of the record elds, which speeds up the

comparison process. Additionally, the induced model can be generated in a reduced amount of

time and is usually characterized by higher predictive accuracy. Verykios et al. [91] also suggest

using a pruning technique on the derived decision trees that are used to classify record pairs as

matches or mismatches. Pruning produces models (trees) of smaller size not only avoid over-

tting and have a higher accuracy, but also allow for faster execution of the matching algorithm.

VI. D UPLICATE D ETECTION T OOLS

Over the past several years, a range of tools for cleaning data has appeared on the market

and research groups have made available to the public software packages that can be used for

duplicate record detection. In this section, we review such packages, focusing on tools that have

open architecture and allow the users to understand the underlying mechanics of the matching

mechanisms.

3
The Febrl system (Freely Extensible Biomedical Record Linkage) is an open-source data

cleaning toolkit, and it has two main components: The rst component deals with data stan-

dardization and the second performs the actual duplicate detection. The data standardization

relies mainly on hidden-Markov models (HMMs); therefore, Febrl typically requires training to

correctly parse the database entries. For duplicate detection, Febrl implements a variety of string

similarity metrics, such as Jaro, edit distance, and q-gram distance (see Section III). Finally,

Febrl supports phonetic encoding (Soundex, NYSIIS, and Double Metaphone) to detect similar

names. Since phonetic similarity is sensitive to errors in the rst letter of a name, Febrl also

3
http://sourceforge.net/projects/febrl

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 32

computes phonetic similarity using the reversed version of the name string, sidestepping the

“rst-letter” sensitivity problem.

TAILOR [30] is a exible record matching toolbox, which allows the users to apply different

duplicate detection methods on the data sets. The exibility of using multiple models is useful

when the users do not know which duplicate detection model will perform most effectively

on their particular data. TAILOR follows a layered design, separating comparison functions

from the duplicate detection logic. Furthermore, the execution strategies, which improve the

efciency are implemented in a separate layer, making the system more extensible than systems

that rely on monolithic designs. Finally, TAILOR reports statistics, such as estimated accuracy

and completeness, which can help the users understand better the quality of the a given duplicate

detection execution over a new data set.

4
WHIRL is a duplicate record detection system available for free for academic and research

use. WHIRL uses the tf.idf token-based similarity metric to identify similar strings within two

5
lists. The Flamingo Project is a similar tools that provides a simple string matching tool that

takes as input two string lists and returns the strings pairs that are within a prespecied edit

distance threshold. WizSame by WizSoft is also a product that allows the discovery of duplicate

records in a database. The matching algorithm is very similar to SoftTF.IDF (see Section III-B):

two records match if they contain a signicant fraction of identical or similar words, where

similar are the words that are within edit distance one.

BigMatch [102] is the duplicate detection program used by the U.S. Census Bureau. It relies

on blocking strategies to identify potential matches between the records of two relations, and

scales well for very large data sets. The only requirement is that one of the two relations should

t in memory, and it is possible to t in memory even relations with 100 million records. The

main goal of BigMatch is not to perform sophisticated duplicate detection, but rather to generate

a set of candidate pairs that should be then processed by more sophisticated duplicate detection

algorithms.

Finally, we should note that currently many database vendors (Oracle, IBM, Microsoft) do not

provide sufcient tools for duplicate record detection. Most of the efforts until now has focused

4
http://www.cs.cmu.edu/˜wcohen/whirl/
5
http://www.ics.uci.edu/˜flamingo/

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 33

on creating easy-to-use ETL tools, that can standardize database records and x minor errors,

mainly in the context of address data. Another typical function of the tools that are provided

today is the ability to use reference tables and standardize the representation of entities that are

well-known to have multiple representations. (For example, “TKDE” is also frequently written

as “IEEE TKDE” or as “Transactions on Knowledge and Data Engineering.”) A recent, positive

step is the existence of multiple data cleaning operators within Microsoft SQL Server Integration

Services, which is part of Microsoft SQL Server 2005. For example, SQL server now includes

the ability to perform “fuzzy matches” and implements “error-tolerable indexes” that allow fast

execution of such approximate lookups. The adopted similarity metric is similar to SoftTF.IDF,

described in Section III-B. Ideally, the other major database vendors would also follow suit and

add similar capabilities and extend the current ETL packages.

VII. F UTURE D IRECTIONS AND C ONCLUSIONS

In this survey, we have presented a comprehensive survey of the existing techniques used

for detecting non-identical duplicate entries in database records. The interested reader may also

want to read a complementary survey by Winkler [100] and the Special Issue of the IEEE Data

Engineering Bulletin on Data Quality [45].

As database systems are becoming more and more commonplace, data cleaning is going to be

the cornerstone for correcting errors in systems which are accumulating vast amounts of errors

on a daily basis. Despite the breadth and depth of the presented techniques, we believe that there

is still room for substantial improvements in the current state-of-the-art.

First of all, it is currently unclear which metrics and techniques are the current state-of-the-

art. The lack of standardized, large scale benchmarking data sets can be a big obstacle for

the further development of the eld, as it is almost impossible to convincingly compare new

techniques with existing ones. A repository of benchmark data sources with known and diverse

characteristics should be made available to developers so they may evaluate their methods during

the development process. Along with benchmark and evaluation data, various systems need

some form of training data to produce the initial matching model. Although small data sets are

available, we are not aware of large-scale, validated data sets that could be used as benchmarks.

Winkler [98] highlights techniques on how to derive data sets that are properly anonymized and

are still useful for duplicate record detection purposes.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 34

Currently, there are two main approaches for duplicate record detection. Research in databases

emphasizes relatively simple and fast duplicate detection techniques, that can be applied to

databases with millions of records. Such techniques typically do not rely on the existence of

training data, and emphasize efciency over effectiveness. On the other hand, research in machine

learning and statistics aims to develop more sophisticated matching techniques that rely on

probabilistic models. An interesting direction for future research is to develop techniques that

combine the best of both worlds.

Most of the duplicate detection systems available today offer various algorithmic approaches

for speeding up the duplicate detection process. The changing nature of the duplicate detection

process also requires adaptive methods that detect different patterns for duplicate detection and

automatically adapt themselves over time. For example, a background process could monitor

the current data, incoming data and any data sources that need to be merged or matched, and

decide, based on the observed errors, whether a revision of the duplicate detection process is

necessary or not. Another related aspect of this challenge is to develop methods that permit the

user to derive the proportions of errors expected in data cleaning projects.

Finally, large amounts of structured information is now derived from unstructured text and

from the web. This information is typically imprecise and noisy; duplicate record detection

techniques are crucial for improving the quality of the extracted data. The increasing popularity

of information extraction techniques is going to make this issue more prevalent in the future,

highlighting the need to develop robust and scalable solutions. This only adds to the sentiment

that more research is needed in the area of duplicate record detection and in the area of data

cleaning and information quality in general.

R EFERENCES

[1] Eugene Agichtein and Venkatesh Ganti. Mining reference tables for automatic text segmentation. In Proceedings of the

Tenth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2004), pages 20–29,

2004.

[2] Rakesh Agrawal and Ramakrishnan Srikant. Searching with numbers. In Proceedings of the 11th International World

Wide Web Conference (WWW11), pages 420–431, 2002.

[3] Ravindra K. Ahuja, Thomas L. Magnanti, and James B. Orlin. Network Flows: Theory, Algorithms, and Applications.

Prentice Hall, 1st edition, February 1993.

[4] Stephen F. Altschula, Warren Gisha, Webb Millerb, Eugene W. Meyersc, and David J. Lipmana. Basic local alignment

search tool. Journal of Molecular Biology, 215(3):403–410, October 1990.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 35

[5] Rohit Ananthakrishna, Surajit Chaudhuri, and Venkatesh Ganti. Eliminating fuzzy duplicates in data warehouses. In

Proceedings of the 28th International Conference on Very Large Databases (VLDB 2002), 2002.

[6] Nikhil Bansal, Avrim Blum, and Shuchi Chawla. Correlation clustering. Machine Learning, 56(1-3):89–113, 2004.

[7] Rohan Baxter, Peter Christen, and Tim Churches. A comparison of fast blocking methods for record linkage. In ACM

SIGKDD '03 Workshop on Data Cleaning, Record Linkage, and Object Consolidation, pages 25–27, 2003.

[8] Indrajit Bhattacharya and Lise Getoor. Latent dirichlet allocation model for entity resolution. Technical Report CS-TR-

4740, Computer Science Department, University of Maryland, August 2005.

[9] Mikhail Bilenko, Raymond J. Mooney, William Weston Cohen, Pradeep Ravikumar, and Stephen E. Fienberg. Adaptive

name matching in information integration. IEEE Intelligent Systems, 18(5):16–23, September/October 2003.

[10] Avrim Blum and Tom Mitchell. Combining labeled and unlabeled data with co-training. In COLT' 98: Proceedings of

the eleventh annual conference on Computational learning theory, pages 92–100, 1998.

[11] Vinayak R. Borkar, Kaustubh Deshmukh, and Sunita Sarawagi. Automatic segmentation of text into structured records.

In Proceedings of the 2001 ACM SIGMOD International Conference on Management of Data (SIGMOD 2001), pages

175–186, 2001.

[12] Leo Breiman, Jerome H. Friedman, Richard A. Olshen, and Charles J. Stone. Classication and Regression Trees. CRC

Press, July 1984.

[13] Andrei Z. Broder, Steven C. Glassman, Mark S. Manasse, and Geoffrey Zweig. Syntactic clustering of the web. In

Proceedings of the Sixth International World Wide Web Conference (WWW6), pages 1157–1166, 1997.

[14] Abhirup Chatterjee and Arie Segev. Data manipulation in heterogeneous databases. ACM SIGMOD Record, 20(4):64–68,

December 1991.

[15] Surajit Chaudhuri, Kris Ganjam, Venkatesh Ganti, and Rajeev Motwani. Robust and efcient fuzzy match for online

data cleaning. In Proceedings of the 2003 ACM SIGMOD International Conference on Management of Data (SIGMOD

2003), pages 313–324, 2003.

[16] Surajit Chaudhuri, Venkatesh Ganti, and Rajeev Motwani. Robust identication of fuzzy duplicates. In Proceedings of

the 21st IEEE International Conference on Data Engineering (ICDE 2005), pages 865–876, 2005.

[17] Peter Cheeseman and John Sturz. Bayesian classication (autoclass): Theory and results. In Advances in knowledge

discovery and data mining, pages 153–180. AAAI Press/The MIT Press, 1996.

[18] Junghoo Cho, Narayanan Shivakumar, and Hector Garcia-Molina. Finding replicated web collections. In Proceedings of

the 2000 ACM SIGMOD International Conference on Management of Data (SIGMOD 2000), pages 355–366, 2000.

[19] Munir Cochinwala, Verghese Kurien, Gail Lalk, and Dennis Shasha. Efcient data reconciliation. Information Sciences,

137(1-4):1–15, September 2001.

[20] William W. Cohen. Data integration using similarity joins and a word-based information representation language. ACM

Transactions on Information Systems, 18(3):288–321, 2000.

[21] William W. Cohen, Henry Kautz, and David McAllester. Hardening soft information sources. In Proceedings of the Sixth

ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2000), pages 255–259, 2000.

[22] William Weston Cohen. Integration of heterogeneous databases without common domains using queries based on textual

similarity. In Proceedings of the 1998 ACM SIGMOD International Conference on Management of Data (SIGMOD'98),

pages 201–212, 1998.

[23] William Weston Cohen and Jacob Richman. Learning to match and cluster large high-dimensional data sets for data

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 36

integration. In Proceedings of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data

Mining (KDD-2002), 2002.

[24] David A. Cohn, Les Atlas, and Richard E. Ladner. Improving generalization with active learning. Machine Learning,

15(2):201–221, 1994.

[25] Tamraparni Dasu, Theodore Johnson, Shanmugauelayut Muthukrishnan, and Vladislav Shkapenyuk. Mining database

structure; or, how to build a data quality browser. In Proceedings of the 2002 ACM SIGMOD International Conference

on Management of Data (SIGMOD 2002), pages 240–251, 2002.

[26] Arthur Pentland Dempster, Nan McKenzie Laird, and Donald Bruce Rubin. Maximum likelihood from incomplete data

via the EM algorithm. Journal of the Royal Statistical Society, B(39):1–38, 1977.

[27] Debabrata Dey, Sumit Sarkar, and Prabuddha De. Entity matching in heterogeneous databases: A distance based decision

model. In 31st Annual Hawaii International Conference on System Sciences (HICSS'98), pages 305–313, 1998.

[28] Nelson S. D'Andrea Du Bois, Jr. A solution to the problem of linking multivariate documents. Journal of the American

Statistical Association, 64(325):163–174, March 1969.

[29] Richard Oswald Duda and Peter Elliot Hart. Pattern Classication and Scene Analysis. Wiley, 1973.

[30] Mohamed G. Elfeky, Ahmed K. Elmagarmid, and Vassilios S. Verykios. TAILOR: A record linkage tool box. In

Proceedings of the 18th IEEE International Conference on Data Engineering (ICDE 2002), pages 17–28, 2002.

[31] Ivan Peter Fellegi and Alan B. Sunter. A theory for record linkage. Journal of the American Statistical Association,

64(328):1183–1210, December 1969.

[32] Helena Galhardas, Daniela Florescu, Dennis Shasha, Eric Simon, and Cristian-Augustin Saita. Declarative data cleaning:

Language, model, and algorithms. In Proceedings of the 27th International Conference on Very Large Databases (VLDB

2001), pages 371–380, 2001.

[33] Leicester E. Gill. OX-LINK: The Oxford medical record linkage system. In Proceedings of the International Record

Linkage Workshop and Exposition, pages 15–33, 1997.

[34] Luis Gravano, Panagiotis G. Ipeirotis, Hosagrahar Visvesvaraya Jagadish, Nick Koudas, Shanmugauelayut Muthukrishnan,

Lauri Pietarinen, and Divesh Srivastava. Using q -grams in a DBMS for approximate string processing. IEEE Data

Engineering Bulletin, 24(4):28–34, December 2001.

[35] Luis Gravano, Panagiotis G. Ipeirotis, Hosagrahar Visvesvaraya Jagadish, Nick Koudas, Shanmugauelayut Muthukrishnan,

and Divesh Srivastava. Approximate string joins in a database (almost) for free. In Proceedings of the 27th International

Conference on Very Large Databases (VLDB 2001), pages 491–500, 2001.

[36] Luis Gravano, Panagiotis G. Ipeirotis, Nick Koudas, and Divesh Srivastava. Text joins in an RDBMS for web data

integration. In Proceedings of the 12th International World Wide Web Conference (WWW12), pages 90–101, 2003.

[37] Sudipto Guha, Nick Koudas, Amit Marathe, and Divesh Srivastava. Merging the results of approximate match operations.

In Proceedings of the 30th International Conference on Very Large Databases (VLDB 2004), pages 636–647, 2004.

[38] Trevor Hastie, Robert Tibshirani, and Jerome Harold Friedman. The Elements of Statistical Learning. Springer Verlag,

August 2001.

[39] Mauricio Antonio Hernández and Salvatore Joseph Stolfo. Real-world data is dirty: Data cleansing and the merge/purge

problem. Data Mining and Knowledge Discovery, 2(1):9–37, January 1998.

[40] Matthew A. Jaro. Unimatch: A record linkage system: User's manual. Technical report, U.S. Bureau of the Census,

Washington, D.C., 1976.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 37

[41] Matthew A. Jaro. Advances in record-linkage methodology as applied to matching the 1985 census of Tampa, Florida.

Journal of the American Statistical Association, 84(406):414–420, June 1989.

[42] Thorsten Joachims. Making large-scale SVM learning practical. In Bernhard Schölkopf, Christopher John C. Burges,

and Alexander J. Smola, editors, Advances in Kernel Methods - Support Vector Learning. MIT-Press, 1999.

[43] Ralph Kimball and Joe Caserta. The data warehouse ETL toolkit: Practical techniques for extracting, cleaning, conforming,

and delivering data. John Wiley & Sons, 2004.

[44] Daphne Koller and Mehran Sahami. Hierarchically classifying documents using very few words. In Proceedings of the

14th International Conference on Machine Learning (ICML'97), pages 170–178, 1997.

[45] Nick Koudas, editor. IEEE Data Engineering Bulletin, volume 29. IEEE, June 2006. Special Issue on Data Quality.

[46] Nick Koudas, Amit Marathe, and Divesh Srivastava. Flexible string matching against large databases in practice. In

Proceedings of the 30th International Conference on Very Large Databases (VLDB 2004), pages 1078–1086, 2004.

[47] Karen Kukich. Techniques for automatically correcting words in text. ACM Computing Surveys, 24(4):377–439, December

1992.

[48] Gad M. Landau and Uzi Vishkin. Fast parallel and serial approximate string matching. Journal of Algorithms, 10(2):157–

169, June 1989.

[49] Vladimir I. Levenshtein. Binary codes capable of correcting deletions, insertions and reversals. Doklady Akademii Nauk

SSSR, 163(4):845–848, 1965. Original in Russian – translation in Soviet Physics Doklady 10(8):707-710, 1966.

[50] Ee-Peng Lim, Jaideep Srivastava, Satya Prabhakar, and James Richardson. Entity identication in database integration.

In Proceedings of the Ninth IEEE International Conference on Data Engineering (ICDE 1993), pages 294–301, 1993.

[51] Nikos Mamoulis. Efcient processing of joins on set-valued attributes. In Proceedings of the 2003 ACM SIGMOD

International Conference on Management of Data (SIGMOD 2003), pages 157–168, 2003.

[52] Andrew McCallum. Information extraction: Distilling structured data from unstructured text. ACM Queue, 3(9):48–57,

2005.

[53] Andrew McCallum, Dayne Freitag, and Fernando C. N. Pereira. Maximum entropy markov models for information

extraction and segmentation. In Proceedings of the 17th International Conference on Machine Learning (ICML 2000),

pages 591–598, 2000.

[54] Andrew McCallum, Kamal Nigam, and Lyle H. Ungar. Efcient clustering of high-dimensional data sets with application

to reference matching. In Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery

and Data Mining (KDD-2000), pages 169–178, 2000.

[55] Andrew McCallum and Ben Wellner. Conditional models of identity uncertainty with application to noun coreference.

In Advances in Neural Information Processing Systems (NIPS 2004), 2004.

[56] Ruslan Mitkov. Anaphora Resolution. Longman, 1st edition, August 2002.

[57] Alvaro E. Monge and Charles P. Elkan. The eld matching problem: Algorithms and applications. In Proceedings of the

Second International Conference on Knowledge Discovery and Data Mining (KDD-96), pages 267–270, 1996.

[58] Alvaro E. Monge and Charles P. Elkan. An efcient domain-independent algorithm for detecting approximately duplicate

database records. In Proceedings of the 2nd ACM SIGMOD Workshop on Research Issues in Data Mining and Knowledge

Discovery (DMKD'97), pages 23–29, 1997.

[59] Gonzalo Navarro. A guided tour to approximate string matching. ACM Computing Surveys, 33(1):31–88, 2001.

[60] Saul Ben Needleman and Christian Dennis Wunsch. A general method applicable to the search for similarities in the

amino acid sequence of two proteins. Journal of Molecular Biology, 48(3):443–453, March 1970.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 38

[61] Howard B. Newcombe. Record linking: The design of efcient systems for linking records into individual and family

histories. American Journal of Human Genetics, 19(3):335–359, May 1967.

[62] Howard B. Newcombe. Handbook of Record Linkage. Oxford University Press, 1988.

[63] Howard B. Newcombe and James M. Kennedy. Record linkage: Making maximum use of the discriminating power of

identifying information. Communications of the ACM, 5(11):563–566, November 1962.

[64] Howard B. Newcombe, James M. Kennedy, S.J. Axford, and A.P. James. Automatic linkage of vital records. Science,

130(3381):954–959, October 1959.

[65] Kamal Nigam, Andrew McCallum, Sebastian Thrun, and Tom M. Mitchell. Text classication from labeled and unlabeled

documents using EM. Machine Learning, 39(2/3):103–134, 2000.

[66] Hanna Pasula, Bhaskara Marthi, Brian Milch, Stuart J. Russell, and Ilya Shpitser. Identity uncertainty and citation

matching. In Advances in Neural Information Processing Systems (NIPS 2002), pages 1401–1408, 2002.

[67] Mike Perkowitz, Robert B. Doorenbos, Oren Etzioni, and Daniel Sabey Weld. Learning to understand information on

the Internet: An example-based approach. Journal of Intelligent Information Systems, 8(2):133–153, March 1997.

[68] Lawrence Philips. Hanging on the metaphone. Computer Language Magazine, 7(12):39–44, December 1990. Accessible

at http://www.cuj.com/documents/s=8038/cuj0006philips/.
[69] Lawrence Philips. The double metaphone search algorithm. C/C++ Users Journal, 18(5), June 2000.

[70] Jose C. Pinheiro and Don X. Sun. Methods for linking and mining heterogeneous databases. In Proceedings of the

International Conference on Knowledge Discovery and Data Mining (KDD-98), pages 309–313, 1998.

[71] Vijayshankar Raman and Joseph M. Hellerstein. Potter's wheel: An interactive data cleaning system. In Proceedings of

the 27th International Conference on Very Large Databases (VLDB 2001), pages 381–390, 2001.

[72] Pradeep Ravikumar and William Weston Cohen. A hierarchical graphical model for record linkage. In 20th Conference

on Uncertainty in Articial Intelligence (UAI 2004), 2004.

[73] Eric Sven Ristad and Peter N. Yianilos. Learning string edit distance. IEEE Transactions on Pattern Analysis and Machine

Intelligence, 20(5):522–532, May 1998.

[74] Elke Rundensteiner, editor. IEEE Data Engineering Bulletin, volume 22. IEEE, January 1999. Special Issue on Data

Transformation.

[75] Robert C. Russell. Index, U.S. patent 1,261,167. Available at http://patft.uspto.gov/netahtml/srchnum.

htm, April 1918.
[76] Robert C. Russell. Index, U.S. patent 1,435,663. Available at http://patft.uspto.gov/netahtml/srchnum.
htm, November 1922.
[77] Sunita Sarawagi, editor. IEEE Data Engineering Bulletin, volume 23. IEEE, December 2000. Special Issue on Data

Cleaning.

[78] Sunita Sarawagi and Anuradha Bhamidipaty. Interactive deduplication using active learning. In Proceedings of the Eighth

ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2002), pages 269–278, 2002.

[79] Sunita Sarawagi and Alok Kirpal. Efcient set joins on similarity predicates. In Proceedings of the 2004 ACM SIGMOD

International Conference on Management of Data (SIGMOD 2004), pages 743–754, 2004.

[80] Parag Singla and Pedro Domingos. Multi-relational record linkage. In KDD-2004 Workshop on Multi-Relational Data

Mining, pages 31–48, 2004.

[81] Temple F. Smith and Michael S. Waterman. Identication of common molecular subsequences. Journal of Molecular

Biology, 147:195–197, 1981.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 39

[82] Aya Soffer, David Carmel, Doron Cohen, Ronald Fagin, Eitan Farchi, Michael Herscovici, and Yoelle S. Maarek. Static

index pruning for information retrieval systems. In Proceedings of the 24th Annual International ACM SIGIR Conference

on Research and Development in Information Retrieval, SIGIR 2001, pages 43–50, 2001.

[83] Erkki Sutinen and Jorma Tarhio. On using q -gram locations in approximate string matching. In Proceedings of Third

Annual European Symposium on Algorithms (ESA'95), pages 327–340, 1995.

[84] Charles Sutton, Khashayar Rohanimanesh, and Andrew McCallum. Dynamic conditional random elds: Factorized

probabilistic models for labeling and segmenting sequence data. In Proceedings of the 21st International Conference on

Machine Learning (ICML 2004), 2004.

[85] Robert L. Taft. Name search techniques. Technical Report Special Report No. 1, New York State Identication and

Intelligence System, Albany, NY, February 1970.

[86] Sheila Tejada, Craig A. Knoblock, and Steven Minton. Learning object identication rules for information integration.

Information Systems, 26(8):607–633, 2001.

[87] Sheila Tejada, Craig Alan Knoblock, and Steven Minton. Learning domain-independent string transformation weights for

high accuracy object identication. In Proceedings of the Eighth ACM SIGKDD International Conference on Knowledge

Discovery and Data Mining (KDD-2002), 2002.

[88] Benjamin J. Tepping. A model for optimum linkage of records. Journal of the American Statistical Association,

63(324):1321–1332, December 1968.

[89] Esko Ukkonen. Approximate string matching with q -grams and maximal matches. Theoretical Computer Science,

92(1):191–211, 1992.

[90] Julian R. Ullmann. A binary n-gram technique for automatic correction of substitution, deletion, insertion and reversal

errors in words. The Computer Journal, 20(2):141–147, 1977.

[91] Vassilios S. Verykios, Ahmed K. Elmagarmid, and Elias N. Houstis. Automating the approximate record matching process.

Information Sciences, 126(1-4):83–98, July 2000.

[92] Vassilios S. Verykios and George V. Moustakides. A generalized cost optimal decision model for record matching. In

Proceedings of the 2004 International Workshop on Information Quality in Information Systems, pages 20–26, 2004.

[93] Vassilios S. Verykios, George V. Moustakides, and Mohamed G. Elfeky. A bayesian decision model for cost optimal

record matching. VLDB Journal, 12(1):28–40, May 2003.

[94] Y. Richard Wang and Stuart E. Madnick. The inter-database instance identication problem in integrating autonomous

systems. In Proceedings of the Fifth IEEE International Conference on Data Engineering (ICDE 1989), pages 46–55,

1989.

[95] Michael S. Waterman, Temple F. Smith, and William A. Beyer. Some biological sequence metrics. Advances in

Mathematics, 20(4):367–387, 1976.

[96] Jennifer Widom. Research problems in data warehousing. In Proceedings of the 1995 ACM Conference on Information

and Knowledge Management (CIKM'95), pages 25–30, 1995.

[97] William E. Winkler. Improved decision rules in the Felligi-Sunter model of record linkage. Technical Report Statistical

Research Report Series RR93/12, U.S. Bureau of the Census, Washington, D.C., 1993.

[98] William E. Winkler. The state of record linkage and current research problems. Technical Report Statistical Research

Report Series RR99/04, U.S. Bureau of the Census, Washington, D.C., 1999.

[99] William E. Winkler. Methods for record linkage and bayesian networks. Technical Report Statistical Research Report

Series RRS2002/05, U.S. Bureau of the Census, Washington, D.C., 2002.

August 13, 2006 DRAFT

AHMED K. ELMAGARMID, PANAGIOTIS G. IPEIROTIS, AND VASSILIOS S. VERYKIOS 40

[100] William E. Winkler. Overview of record linkage and current research directions. Technical Report Statistical Research

Report Series RRS2006/02, U.S. Bureau of the Census, Washington, D.C., 2006.

[101] William E. Winkler and Yves Thibaudeau. An application of the Fellegi-Sunter model of record linkage to the 1990 U.S.

decennial census. Technical Report Statistical Research Report Series RR91/09, U.S. Bureau of the Census, Washington,

D.C., 1991.

[102] William E. Yancey. Bigmatch: A program for extracting probable matches from a large le for record linkage. Technical

Report Statistical Research Report Series RRC2002/01, U.S. Bureau of the Census, Washington, D.C., March 2002.

[103] William E. Yancey. Evaluating string comparator performance for record linkage. Technical Report Statistical Research

Report Series RRS2005/05, U.S. Bureau of the Census, Washington, D.C., June 2005.

[104] Justin Zobel, Alistair Moffat, and Kotagiri Ramamohanarao. Inverted les versus signature les for text indexing. ACM

Transactions on Database Systems, 23(4):453–490, December 1998.

August 13, 2006 DRAFT