what-when-how
In Depth Tutorials and Information
subsequently at node
v
, and let
S
2
denote the set of topics
j
such that
u
was infected
with topic
j
but
v
was never infected with the topic. For each topic
j
∈
S
1
, require as
input the pair (
p
j
,
δ
j
), where
p
j
is the posterior probability computed earlier, and
δ
j
is
the delay in days between the appearance of the topic in
u
and in
v
. For every topic
j
∈
S
2
, require as input the value
δ
j
, where
δ
j
days elapsed between the appearance
of topic
j
at node
u
and the end of the snapshot. hen estimate an updated version
of
r
and
k
as follows:
∑
∑
∑
∑
p
p
j
j
j S
∈
j S
∈
=
(3.15)
r
:
=
k
:
1
1
≤
δ
p
δ
Pr[
r
]
j
j
j
∈
∪
j S
∈
j S
S
1
2
1
where
P a
[
≤
b
]
=
(
1
−
a
)(
1
− −
(
1
a
) )
is the probability that a geometric distribution
b
r
with parameter
a
has value ≤
b
.
Now take an improved guess at the transmission graph, so return to the soft-
assignment step and recompute the posteriors, iterating until convergence. In the
first step, use the model of the graph to guess how data traveled; in the second, use
the guess about how data traveled to improve the model of the graph.
Some investigators have improved the work we have just described. Leskovec
et al. [32] find that the popularity of posts drops with a power law, and the size
distribution of cascades follows a perfect Zipfian distribution; based on these,
they present a simple model that mimics the spread of information on the blo-
gosphere and produces information cascades very similar to those in real life.
Kleinberg [33] considers a collection of probabilistic and game-theoretic mod-
els for information cascades through the network and investigates the cascading
behavior in a number of online settings, including word-of-mouth effects in the
success of new products and the influence of social networks in the growth of
online communities.
3.2.2.5 Analysis of Users' Interests
As described in the initial part of section 2.2 in chapter 2, mining users' interests
can be the basis of analysis and modeling of their reading and posting behavior,
and supply more inspiration for this work. Generally, mining users' interests is to
discover which type of topics can raise one user's interest, or how to display the
information that may attract more users.
Teng and Chen [34] proposed a method to detect bloggers' interest from three
kinds of important features (textual features, temporal features, and interactive fea-
tures) contained in blogs. he analysis of textual features comprises three aspects.
First, examine the interest words relative to the all words used in a weblog. hen
observe the number of interest documents relative to the overall entries in a weblog.
