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

5.4 Supervised Learning 115

Table 5.1. A Sample Dataset. In this dataset, features are

characteristics of individuals on Twitter, and the class attribute

denotes whether they are influential or not

ID Celebrity Verified Account # Followers Influential?

1 Yes No 1.25 M No

2 No Yes 1 M No

3 No Yes 600 K No

4 Yes Unknown 2.2 M No

5 No No 850 K Yes

6 No Yes 750 K No

7 No No 900 K Yes

8 No No 700 K No

9 Yes Yes 1.2 M No

10 No Unknown 950 K Yes

5.4.1 Decision Tree Learning

Consider the dataset shown in Table5.1. The last attribute represents the

class attribute, and the other attributes represent the features. In decision tree

classification, a decision tree is learned from the training dataset, and that

tree is later used to predict the class attribute value for instances in the test

dataset. As an example, two learned decision trees from the dataset shown

in Table5.1are provided in Figure5.3. As shown in this figure, multiple

decision trees can be learned from the same dataset, and these decision trees

can both correctly predict the class attribute values for all instances in the

dataset. Construction of decision trees is based on heuristics, as different

heuristics generate different decision trees from the same dataset.

Splitting Attributes

Ye s Celebrity N o

Celebrity

No AccountVerified

Verified

Account

No, Unknown

No, Unknown

Number of

Followers

Number of

<800 K >800 K <800 K Followers >800 K

No

No

No

No

No

Ye s No

Ye s

Ye s

Ye s

Ye s

(a) Learned Decision Tree 1 (b) Learned Decision Tree 2

Figure 5.3. Decision Trees Learned from Data Provided in Table5.1.