People-LDA: Anchoring Topics to People using Face Recognition

This page is intended to display more results which could not be accomodated in the following paper due to space constraints. We also intend to share the code for the model proposed in our paper:

Vidit Jain, Erik Learned-Miller, Andrew McCallum.
People-LDA: Anchoring Topics to People using Face Recognition.
International Conference on Computer Vision (ICCV), 2007, Rio De Janeiro, Brazil.

Good cluster examples Noisy cluster examples

Examples of good people-topics and corresponding distributions of words

Note that some clusters have more than one people that are similar to each other visually (see Kofi Annan's topic) or through some underlying theme of the topic (Parvez Musharraf and Benazir Bhutto in Parvez Musharraf's topic).
aniston
los
angeles
hollywood
actress
jennifer
emmy
awards
annual
series
security
homeland
department
ridge
washington
police
director
tom
reporters
john
annan
williams
general
kofi
chief
secretary
nations
file
iraq
blix
musharraf
conference
general
pakistan
pervez
north
islamabad
korea
war
bhutto
united
states
iraq
mahathir
photo
general
told
mohamad
attack
inspectors
daschle
leader
senate
tom
majority
bush
white
house
lott
trent



Examples of noisy clusters and corresponding distributions of words


Note that some topics did not emerge around people but represent a group (tennis, in the top-left topic), or contain many face images of a different person with a similar appearance (Jennifer Aniston in the Britney Spears' topic and Vladimir Putin in the Igor Ivanov's topic).
tournament
returns
match
tennis
won
hewitt
blake
masters
australia
ball
seen
file
ring
arrives
shown
photo
charged
case
step
store
spears
film
city
star
premiere
poses
britney
watts
mexico
week
open
york
states
sampras
flushing
united
agassi
defeated
gambill
set
bush
ivanov
house
media
white
igor
meeting
defense
russian
george
weapons
blix
iraq
resolution
nations
council
united
inspector
inspectors
chief