what-when-how
In Depth Tutorials and Information
C j
p
C j
Commentdistributionvector
( )
: his vector can be represented as
( )
= <
C j
z
p
z
( )
= <
c
,
c
,
c
,...,
c
>
,
1 2 3
where
c
1j
represents the distribution of comment
on topic 1 for blogger
j
within time window
z
. Here the weight of
c
1
j
is calculated as
the percentage of comments belonging to topic 1 (denoted as |
c
1
j
|) posted by blog-
ger
j
against the total number of comments posted by blogger
j
(denoted as |
cj
|) in
the time window
z
.
Socialnetworkvector
(
p
z
j
j
j
nj
z
S j
z
: he social network features of blogger
j
in time
window
z
are represented as a vector
S j
( ) )
( )
= <
s
,
s
,
s
,...,
s
>
in
z
1
j
2
j
3
j
nj
z
S j
m
m
C
TC
∑
∑
j
→
x
=
⋅
=
(3.13)
( )
T x TC
( ) ,
C
z
p
z
j
j
→
x
j
=
=
x
1
x
1
where
m
is the total number of social neighbors of blogger
j
in the network,
C
j
→
x
represents the number of comments written by blogger
j
to blog entries posted by
blogger
x
in a certain time window, and
TC
j
represents the total number of com-
ments written by blogger
j
in the same time window.
Based on the social-network and profile-based blogging behavior features
<
>
, ( ) , ( ) , ( )
for blogger
j
, we can train the social-network and
profile-based blogging-behavior model and predict the future blogging behaviors of
blogger
j
by using regression techniques. We take the previous
k
combined vectors
<
T T j C j
S j
z
p
z
p
z
z
, ( ) , ( ) , ( )
from the (
z
-
k
+ 1)-th time window to the
i
-th time
window as the input vectors, and the combined vector
<
T T j C j
S j
>
z
p
z
p
z
z
>
, ( ) , ( ) , ( )
in the (
z
+ 1)-th time window as the target vector to train the model. hen, by using
the trained regression model, the future blogging behavior of blogger
j
can be pre-
dicted based on the historical general blogging behavior, the blogger's own histori-
cal blogging behavior, and his or her neighbors' historical blogging behavior.
T T j C j
S j
z
p
z
p
z
z
3.2.2.4 Modeling the Posting Behavior Based
on the Cascade Model
In the traditional research on modeling, the spread of an idea or innovation
throughout a social network
G
can be represented as a direct graph. here are two
basic models: the independent cascade model [48] and the linear threshold model
[49]. In the innovation-spreading process, one user's states can be divided into
active (an adopter of the innovation) or inactive. herefore, the aim of modeling
the innovation-spreading process can be transferred to predicting the user node's
tendency to become active, which increases monotonically as more of its neighbors
become active. hus, the process will look roughly as follows from the perspective
of an initially inactive node
u
: as time unfolds, more and more of
u
's neighbors
become active; at some point, this may cause
u
to become active, and
u
's decision
may in turn trigger further decisions by nodes to which
u
is connected.
