COMP3430 / COMP8430 Data wrangling
Lab 5: Classification for Record Linkage
Objectives of this lab
● Today’s lab is the third in a series of five labs during which we will gradually build a complete record linkage system.
● We will be working with different classification techniques and learn how they work and why they are important in the RL process.
● Completion of the classification module in the program.
Outline of this lab
● Learn how different classification techniques work ● Implement different classification techniques
● Evaluate different classification techniques
● Summary
Preliminaries
● Before you begin, aim to review lectures 17 and 18 if you have not already viewed them.
● Go back over the work from lab 4 and remind yourself what we were doing and how the overall program is structured.
● You can download the comparison module with sample solutions in week 7 and use with your RL program if you find difficulties implementing the required comparison functions.
What is Classification?
● This week we focus on the next step in the linkage process, classification.
● The aim of a classification technique is to classify a given pair of records as either a match, a non-match, or a potential match.
● Why do you think we need different classification techniques other than exact classification?
How to classify record pairs
● Before we begin let us see how different classification techniques work. The classification techniques are outlined in lectures 17 and 18.
● Let us assume we have the following two vectors of similarities. – record pair (r1,r2) resulted in: [0.5, 0.9, 0.2, 0.8, 1.0, 0.0, 0.7] – record pair (r3,r4) resulted in: [0.8, 0.7, 1.0, 0.9, 0.6, 0.7, 0.9]
● See if you can compute the classification outcomes of above record pairs using:
– Threshold based classification with thresholds 0.5 and 0.7
– Minimum threshold based classification with thresholds 0.5 and 0.7
How to classify record pairs
Similarity vector for record pair (r1,r2): [0.5, 0.9, 0.2, 0.8, 1.0, 0.0, 0.7]
● Threshold based classification with the threshold 0.5
Average similarity = (0.5 + 0.9 + 0.2 + 0.8 + 1.0 + 0.0 + 0.7) / 7 = 0.5857
Since this average similarity > 0.5, the record pair (r1,r2) is classified as a match.
● Minimum threshold based classification with the threshold 0.5
Chech if each similarity value in the similarity vector is at least 0.5.
Since some similarties are not at least 0.5, the record pair (r1,r2) is classified as a non-match.
Implement different classification techniques
● Now start looking at classification.py and explore how the classification techniques work (inputs, return values, etc.).
● We have already provided a classification technique, exactClassify().
● Run the RL program using this classification technique and see what the
output looks like and how it performs.
● Now try to implement the other classification functions as required in the lab tutorial document.
Questions to consider
● How many matches do you find with each classification technique for the same threshold value?
● Do different thresholds have an effect on the number of matches?
● How do different weights for the weighted average function influence the
number of matches?
● Extra tasks – see if you can change the program to learn the weights for attributes based on their value distributions.
Summary
● In this lab we implemented different classification techniques and learnt how they can be used in the RL program.
● Make sure to complete any unfinished work in this module before you come to the next lab.
● In the next lab we will be looking at how different evaluation measures work and how they can be used in the RL program to evaluate performance.