Home » Java » Core Java » Record Linkage: Playing around with Duke

About Mark Needham

Record Linkage: Playing around with Duke

I’ve become quite interesting in record linkage recently and came across the Duke project which provides some tools to help solve this problem. I thought I’d give it a try.

The typical problem when doing record linkage is that we have two records from different data sets which represent the same entity but don’t have a common key that we can use to merge them together. We therefore need to come up with a heuristic that will allow us to do so.

Duke has a few examples showing it in action and I decided to go with the linking countries one. Here we have countries from Dbpedia and the Mondial database and we want to link them together.

The first thing we need to do is build the project:

export JAVA_HOME=`/usr/libexec/java_home`
mvn clean package -DskipTests

At the time of writing this will put a zip fail containing everything we need at duke-dist/target/. Let’s unpack that:

unzip duke-dist/target/duke-dist-1.3-SNAPSHOT-bin.zip

Next we need to download the data files and Duke configuration file:

wget https://raw.githubusercontent.com/larsga/Duke/master/doc/example-data/countries-dbpedia.csv
wget https://raw.githubusercontent.com/larsga/Duke/master/doc/example-data/countries.xml
wget https://raw.githubusercontent.com/larsga/Duke/master/doc/example-data/countries-mondial.csv
wget https://raw.githubusercontent.com/larsga/Duke/master/doc/example-data/countries-test.txt

Now we’re ready to give it a go:

java -cp "duke-dist-1.3-SNAPSHOT/lib/*" no.priv.garshol.duke.Duke --testfile=countries-test.txt --testdebug --showmatches countries.xml
ID: '7706', NAME: 'guatemala', AREA: '108890', CAPITAL: 'guatemala city',
MATCH 0.9825124555160142
ID: '10052', NAME: 'pitcairn islands', AREA: '47', CAPITAL: 'adamstown',
ID: 'http://dbpedia.org/resource/Pitcairn_Islands', NAME: 'pitcairn islands', AREA: '47', CAPITAL: 'adamstown',
Correct links found: 200 / 218 (91.7%)
Wrong links found: 0 / 24 (0.0%)
Unknown links found: 0
Percent of links correct 100.0%, wrong 0.0%, unknown 0.0%
Records with no link: 18
Precision 100.0%, recall 91.74311926605505%, f-number 0.9569377990430622

We can look in countries.xml to see how the similarity between records is being calculated:


So we’re working out similarity of the capital city and country by calculating their Levenshtein distance i.e. the minimum number of single-character edits required to change one word into the other

This works very well if there is a typo or difference in spelling in one of the data sets. However, I was curious what would happen if the country had two completely different names e.g Cote d’Ivoire is sometimes know as Ivory Coast. Let’s try changing the country name in one of the files:

"19147","Cote dIvoire","Yamoussoukro","322460"
java -cp "duke-dist-1.3-SNAPSHOT/lib/*" no.priv.garshol.duke.Duke --testfile=countries-test.txt --testdebug --showmatches countries.xml
ID: '19147', NAME: 'ivory coast', AREA: '322460', CAPITAL: 'yamoussoukro',

I also tried it out with the BBC and ESPN match reports of the Man Utd vs Tottenham match – the BBC references players by surname, while ESPN has their full names.

When I compared the full name against surname using the Levenshtein comparator there were no matches as you’d expect. I had to split the ESPN names up into first name and surname to get the linking to work.

Equally when I varied the team name’s to be ‘Man Utd’ rather than ‘Manchester United’ and ‘Tottenham’ rather than ‘Tottenham Hotspur’ that didn’t work either.

I think I probably need to write a domain specific comparator but I’m also curious whether I could come up with a bunch of training examples and then train a model to detect what makes two records similar. It’d be less deterministic but perhaps more robust.

Reference: Record Linkage: Playing around with Duke from our JCG partner Mark Needham at the Mark Needham Blog blog.

Do you want to know how to develop your skillset to become a Java Rockstar?

Subscribe to our newsletter to start Rocking right now!

To get you started we give you our best selling eBooks for FREE!


1. JPA Mini Book

2. JVM Troubleshooting Guide

3. JUnit Tutorial for Unit Testing

4. Java Annotations Tutorial

5. Java Interview Questions

6. Spring Interview Questions

7. Android UI Design


and many more ....


Receive Java & Developer job alerts in your Area

I have read and agree to the terms & conditions


Notify of

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Inline Feedbacks
View all comments