Intelligent Database Systems Lab 國國國國國國國國 National Yunlin University of Science and Technology 1 Modeling Semantic Similarities in Multiple Maps Presenter : Wei-Hao Huang Authors : Laurens van der Maaten, Geoffrey Hinton EWI-ICT TR, 2009
Feb 23, 2016
Intelligent Database Systems Lab
國立雲林科技大學National Yunlin University of Science and Technology
1
Modeling Semantic Similarities inMultiple Maps
Presenter : Wei-Hao Huang Authors : Laurens van der Maaten, Geoffrey Hinton
EWI-ICT TR, 2009
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
2
Outlines Motivation Objectives Methodology Experiments Conclusions Comments
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
3
Motivation· Semantic space models cannot faithfully
represent intransitive pairwise similarities or the similarities of words that have multiple meanings.─ Triangle inequality─ Nearest neighbor is limited─ Similarities are symmetric
tie
suittuxedo
ropeknot
Animal
dogdog
dog
dog
dogChina
North Korea
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Objectives
4
• To propose multiple map SNE to solve fundamental limitations of metric spaces
tie
suittuxedo
ropeknot
tie
suittuxedo
tie
rope
knot
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
5
Methodology· Stochastic neighbor embedding· Multiple maps SNE
Map Map2Map1 Map3
Data
SNE Multiple maps SNE
Data
Mixing proportion
(importance)
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology· Stochastic neighbor embedding
6
tie
suittuxedo
ropeknot
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology· Multiple map SNE
7
tieropesuit
tie animal animal
dog dog
dog
dog dog
dog
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology· Multiple map SNE
8
A*C=1*1/2=1/2
B*C=1*1/2=1/2
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Visualization Experiments
─ Florida State University word association dataset─ Selecting 5019 words
· Generalization Experiments─ To evidence their model for semantic representation─ Training data: 80%─ Validation data: 10%─ Test data: 10%
9
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Visualization
10
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments
11
Sport
Clothing
Statue of Liberty
Cheerleader
Cheerleader
Tie
Tiemonarchy
monarchy
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Generalization
12
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Comparing multiple maps SNE with other
method.─ Semantic space models─ Semantic networks ─ Topic models
13
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Conclusions· The multiple maps SNE alleviates the
fundamental limitations of metric spaces.· Multiple map model has characteristics that
are similar to those of topic models.
14
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
15
Comments· Advantages
─ Multiple maps SNE alleviates the fundamental limitations of metric spaces
· Applications─ Data visualization─ Semantic similarities