List of Figures
xv
9.4
Comparing error rates, two classes, same size, same number of
labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
9.5
Comparing error rates, four classes, same size, same number of
labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.6
Comparing error rates for two classes, completely balanced . . . 132
9.7
Comparing error rates, differing numbers of labelled records . . . 133
9.8
Boxplots of error rates . . . . . . . . . . . . . . . . . . . . . . . 134
9.9
Comparing error rates, different class sizes . . . . . . . . . . . . 135
9.10
Comparing error rates, two classes, different sizes, same number
of labelled nodes . . . . . . . . . . . . . . . . . . . . . . . . . . 136
10.1
Illustration of our typed directed connection model . . . . . . . . 140
10.2
Directed Sampson network with positive edges . . . . . . . . . . 142
10.3
Directed Sampson network with negative edges . . . . . . . . . . 143
10.4
Overall relationships among Algerian groups . . . . . . . . . . . 145
10.5
Negative relationships among Algerian groups . . . . . . . . . . 146
10.6
Positive relationships among Algerian groups . . . . . . . . . . . 147
10.7
Negative relationships among Libyan groups . . . . . . . . . . . 148
10.8
Zoomed-in relationships among Libyan groups . . . . . . . . . . 148
10.9
Positive relationships among Libyan groups . . . . . . . . . . . . 149
10.10 Negative relationships among Nigerian groups . . . . . . . . . . 150
10.11 Positive relationships among Nigerian groups . . . . . . . . . . . 151
10.12 Radical groups embedding with 173 groups . . . . . . . . . . . . 151
10.13 Positive relationships of radical groups embedding with 111 groups 152
10.14 Negative relationships of radical groups embedding with 111 groups154
10.15 Positive relationships of radical groups embedding with 16 groups 154
10.16 Negative relationships of radical groups embedding with 16 groups 155
10.17 Positive relationships of violent groups embedding with 65 groups 155
10.18 Negative relationships of violent groups embedding with 65 groups 156
A.1
A cut with two copies of a node in two different clusters . . . . . 163
xv
9.4
Comparing error rates, two classes, same size, same number of
labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
9.5
Comparing error rates, four classes, same size, same number of
labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.6
Comparing error rates for two classes, completely balanced . . . 132
9.7
Comparing error rates, differing numbers of labelled records . . . 133
9.8
Boxplots of error rates . . . . . . . . . . . . . . . . . . . . . . . 134
9.9
Comparing error rates, different class sizes . . . . . . . . . . . . 135
9.10
Comparing error rates, two classes, different sizes, same number
of labelled nodes . . . . . . . . . . . . . . . . . . . . . . . . . . 136
10.1
Illustration of our typed directed connection model . . . . . . . . 140
10.2
Directed Sampson network with positive edges . . . . . . . . . . 142
10.3
Directed Sampson network with negative edges . . . . . . . . . . 143
10.4
Overall relationships among Algerian groups . . . . . . . . . . . 145
10.5
Negative relationships among Algerian groups . . . . . . . . . . 146
10.6
Positive relationships among Algerian groups . . . . . . . . . . . 147
10.7
Negative relationships among Libyan groups . . . . . . . . . . . 148
10.8
Zoomed-in relationships among Libyan groups . . . . . . . . . . 148
10.9
Positive relationships among Libyan groups . . . . . . . . . . . . 149
10.10 Negative relationships among Nigerian groups . . . . . . . . . . 150
10.11 Positive relationships among Nigerian groups . . . . . . . . . . . 151
10.12 Radical groups embedding with 173 groups . . . . . . . . . . . . 151
10.13 Positive relationships of radical groups embedding with 111 groups 152
10.14 Negative relationships of radical groups embedding with 111 groups154
10.15 Positive relationships of radical groups embedding with 16 groups 154
10.16 Negative relationships of radical groups embedding with 16 groups 155
10.17 Positive relationships of violent groups embedding with 65 groups 155
10.18 Negative relationships of violent groups embedding with 65 groups 156
A.1
A cut with two copies of a node in two different clusters . . . . . 163