Information Technology Reference
In-Depth Information
Table 5.3
Visualization of discovered YouTube topics
Word
gameplay xbox playstation gaming minecraft
“Epic Mods - MW2 MOD IN CoD4”
“HEXXIT COOP ep7 w/ Double”
Topic
#1
Video
“Halo 4 Adrift Multiplayer Map”
Word
history german berlin germany poetry
“GEH STERBEN, DU OPFER!!!”
“Syrien - Wahrheit ber das Massaker”
Topic
#17
Video
“Vo l k e r Pispers - Einzeltater”
u
Twi
1
u
Twi
.
u
Twi
{
,...,
|
U
|
}
represents the Twitter followee topic distribution for user
u
. Table
5.2
shows three out of the discovered 80 Twitter followee topics. Each
topic is represented by its top-3 followees and the followees' profile information.
It is shown that the discovered Twitter followee topics have a quite wide coverage,
including game-related general topic #43, Forbes influencer specific topic #10 and
the geographic topic #38.
At YouTube side, for each overlapped user, we crawl the videos he/she uploads,
favorites or adds-to-playlist, to obtain totally 2M YouTube videos. Since YouTube
video topics are expected to span over both textual and visual spaces, for each video,
we collect its textual metadata and visual keyframes.We employ themultimodal topic
model, Corr-LDA [
4
], to obtain theYouTube video topics and video topic distribution.
By direct aggregation, we can get users' YouTube video topic distribution
U
Yo u
∈
U
=
u
Yo u
1
u
Yo u
{
,...,
|
U
|
}
. Table
5.3
shows two out of the discovered 40YouTube video topics,
represented by top-5 probable words and three most representative videos.
