Social Media Mining: An Introduction

P1: Sqe Trim: 6.125in×9.25in Top: 0.5in Gutter: 0.75in
CUUS2079-05 CUUS2079-Zafarani 978 1 107 01885 3 January 13, 2014 19:23

5.1 Data 107

Attributes Class

Name Money Spent Bought Similar Visits Will Buy

John High Yes Frequently?

Mary High Yes Rarely Yes

A dataset is represented using a set offeatures, and an instance is rep-

INSTANCE,

POINT,

DATA POINT,

OR

OBSERVATION

resented using values assigned to these features. Features are also known

asmeasurementsorattributes. In this example, the features areName, FEATURES,

MEASUREMENTS,

OR

ATTRIBUTES

Money Spent,Bought Similar, andVisits; feature values for the

first instance areJohn,High,Yes, andFrequently. Given the feature

values for one instance, one tries to predict itsclass(orclass attribute)

value. In our example, the class attribute isWill Buy, and our class value

prediction for first instance isYes. An instance such as John in which the

class attribute value is unknown is called anunlabeledinstance. Similarly, a

labeledinstance is an instance in which the class attribute value in known. LABELED

AND

UNLABELED

Mary in this dataset represents a labeled instance. The class attribute is

optional in a dataset and is only necessary for prediction purposes. One

can have a dataset in which no class attribute is present, such as a list of

customers and their characteristics.

There are different types of features based on the characteristics of the

feature and the values they can take. For instance,Money Spentcan be

represented using numeric values, such as$25. In that case, we have a

continuous feature, whereas in our example it is adiscretefeature, which

can take a number of ordered values:{High, Normal, Low}.

Different types of features were first introduced by psychologist Stanley

SmithStevens [1996] as “levels of measurement” in the theory of scales. LEVELS OF

He claimed that there are four types of features. For each feature type, there MEASUREMENT

exists a set of permissible operations (statistics) using the feature values

and transformations that are allowed.

 Nominal (categorical). These features take values that are often rep-

resented as strings. For instance, a customer’s name is a nominal

feature. In general, a few statistics can be computed on nominal fea-

tures. Examples are the chi-square statistic (χ^2 ) and themode(most

common feature value). For example, one can find the most com-

mon first name among customers. The only possible transformation

on the data is comparison. For example, we can check whether our

customer’s name is John or not. Nominal feature values are often

presented in a set format.