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

114 Data Mining Essentials

Learning

Algorithm

Training

Set

Learn

Model

Apply

Model Model

Deduction

Test Set

Induction

Figure 5.2. Supervised Learning.

this set are tuples in the format (x,y), wherexis a vector andyis the class

attribute, commonly a scalar. Supervised learning builds a model that maps

xtoy. Roughly, our task is to find a mappingm(.) such thatm(x)=y.We

are also given an unlabeled dataset ortestdataset, in which instances are in

the form (x,?) andyvalues are unknown. Givenm(.) learned from training

data andxof an unlabeled instance, we can computem(x), the result of

which is prediction of the label for the unlabeled instance.

Consider the task of detecting spam emails. A set of emails is given where

users have manually identified spam versus non-spam (training data). Our

task is to use a set of features such as words in the email (x) to identify

the spam/non-spam status (y) of unlabeled emails (test data). In this case,

y={spam,non-spam}.

Supervised learning can be divided intoclassificationandregression.

When the class attribute is discrete, it is called classification; when the

class attribute is continuous, it is regression. We introduce classification

methods such asdecision tree learning,naive Bayes classifier,k-nearest

neighbor classifier, andclassification with network informationand regres-

sion methods such aslinear regressionandlogistic regression. We also

introduce how supervised learning algorithms are evaluated. Before we

delve into supervised learning techniques, we briefly discuss the systematic

process of a supervised learning algorithm.

This process is depicted in Figure5.2. It starts with a training set (i.e.,

labeled data) where both features and labels (class attribute values) are

known. A supervised learning algorithm is run on the training set in a pro-

cess known asinduction. In the induction process, themodelis generated.

The model maps the feature values to the class attribute values. The model

is used on atest setin which the class attribute value is unknown to predict

these unknown class attribute values (deductionprocess).