SlideShare a Scribd company logo
Daesik Kim*, Seonhoon Kim*, Nojun Kwak
김대식
Textbook QA?
The most realistic QA problem !
Textbook QA?
The most complicated QA problem !
Context
Question
Input
Answer
Output
Question
Answer
Input
Context
Output
https://tenor.com/view/johnny-johnny5-johnnyfive-reading-scan-gif-12961036
Skimming step
… and the title is …
Q What is the title of this paper?
https://media-assets.bookbub.com/wp-content/uploads/
2015/12/cute-baby-skim-reading-gif.gif
Context
Solving step
PROBLEM
Q Plates move over Earth’s surface
because of _________
a) conduction within the crust.
b) radiation from the inner core.
c) subduction in the outer core.
d) convection within the mantle.
http://www.grygla.k12.mn.us/uploads/2/3/7/1/23718473/earth_sci_-_ch._6_-_plate_tectonics.pdf
Context
Q Plates move over Earth’s surface
because of _________
a) conduction within the crust.
b) radiation from the inner core.
c) subduction in the outer core.
d) convection within the mantle.
Q What is the southern most point
of the ring of fire?
a) South sandwich trench
b) Japan trench
c) Aleutian trench
d) Kurile trench
http://www.grygla.k12.mn.us/uploads/2/3/7/1/23718473/earth_sci_-_ch._6_-_plate_tectonics.pdf
(Kembhavi et al., 2017)
single sentence / multiple sentences within a paragraph
question/context diagram, and one paragraph
METHODS
Overall Framework
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
…
Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Q & a#1 & C#2 Correct? Wrong?
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Top-1
Tf-idf score
…
model
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Top-1
Tf-idf score
…
Q & a#1 & C#2 Correct? Wrong?
Q & a#2 & C#1 Correct? Wrong?
model
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Top-1
Tf-idf score
…
Q & a#1 & C#2 Correct? Wrong?
Q & a#2 & C#1 Correct? Wrong?
Q & a#3 & C#2 Correct? Wrong?
model
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Top-1
Tf-idf score
…
Q & a#1 & C#2 Correct? Wrong?
Q & a#2 & C#1 Correct? Wrong?
Q & a#3 & C#2 Correct? Wrong?
Q & a#4 & C#9 Correct? Wrong?
model
Self-supervised comprehension
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
… Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Top-1
Tf-idf score
Q & a#1 → C#2 (Top-1 tf-idf score)
→ C#4 (Top-2 tf-idf score)
→ C#1 (Top-3 tf-idf score)
→ C#7 (Top-4 tf-idf score)
→ C#5 (Top-5 tf-idf score)
Q & a#1 & C#2 Correct
Q & a#1 & C#4 Wrong
Q & a#1 & C#1 Wrong
Q & a#1 & C#7 Wrong
Q & a#1 & C#5 Wrong
Self-supervised comprehension
Context
Paragraph #1
Paragraph #2
Paragraph #3
Paragraph #9
Paragraph #10
Q What is the question?
a) answer option #1
b) answer option #2
c) answer option #3
d) answer option #4
Q & a#1 & C#2 Correct? Wrong?
Q & a#1 → C#2 (Top-1 tf-idf score)
→ C#4 (Top-2 tf-idf score)
→ C#1 (Top-3 tf-idf score)
→ C#7 (Top-4 tf-idf score)
→ C#5 (Top-5 tf-idf score)
Top-1
Tf-idf score…
m
model
Self-supervised comprehension
Multi-modal context graph understanding: visual context
rabbit
connects to
fox
mouse
ladybird
snake
14 objects
22 stages
…
diagram parsing from UDPnet
(Kim et al., 2018)
parsed info from UDPnet
& OCR info from TQA
build context graph
of diagrams
rabbit
fox
mouseladybird
14
objects
22
stages
context matrix 𝑪 𝒅
adjacency matrix 𝑨 𝒅
Multi-modal context graph understanding: textual context
… and the title is …
Context
Q What is the title of this paper?
anchor node
dependency
parsing
filtering by
anchor
nodes
context matrix 𝑪 𝒕
adjacency matrix 𝑨𝒕
3
1
1
1
1
2
2
Multi-modal context graph understanding
Question
Answer
word representation
(word embedding;
char representation;
anchor flag)
BiLSTM
Q repr. : 𝒉 𝒒
A repr. : 𝒉 𝒂
visual Context graph
textual Context graph
GCN
GCN GCN
attention
fused representation (f-GCN1) fused representation (f-GCN2)
…
max pooling
𝑯 𝒄
𝟏
𝑯 𝒄
𝟐
𝑨 𝒅
𝑨 𝒕
𝑨 𝒕
𝑨 𝒅
, 𝑨 𝒕
: adjacency matrix
Question & Answer & Context Correct? Wrong?model
Multi-modal problem solving
𝑘𝑡ℎ input for the solving RNN
𝒄𝒐𝒏𝒄𝒂𝒕(𝒉 𝒒; 𝒉 𝒂; 𝑨𝒕𝒕 𝒒; 𝑨𝒕𝒕 𝒂; 𝑨𝒕𝒕 𝒒
𝒒𝒅
; 𝑨𝒕𝒕 𝒂
𝒒𝒅
)
BiLSTM
FC
ෝ𝒚 𝟏 ෝ𝒚 𝒌…
𝒌 𝒕𝒉
𝑨𝒕𝒕 𝒒 = 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒒, 𝑯 𝒄)
𝑨𝒕𝒕 𝒂 = 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒂, 𝑯 𝒄)
𝑨𝒕𝒕 𝒒
𝒒𝒅
= 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒒, 𝑯 𝒒𝒅)
𝑨𝒕𝒕 𝒂
𝒒𝒅
= 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒂, 𝑯 𝒒𝒅)
only for the diagram question
RESULTS
Experiments: quantitative results
Model Text T/F Text MC Text All Diagram All
Random 50.10 22.88 33.62 24.96 29.08
MemN+VQA (Kembhavi et al., 2017) 50.50 31.05 38.73 31.82 35.11
MemN+DPG (Kembhavi et al., 2017) 50.50 30.98 38.69 32.83 35.62
BiDAF+DPG (Kembhavi et al., 2017) 50.40 30.46 38.33 32.72 35.39
Challenge - - 45.57 35.85 40.48
IGMN (Li et al., 2018) 57.41 40.00 46.88 36.35 41.36
Our full model w/o visual context 62.32 49.15 54.35 36.61 45.06
Our full model w/ f-GCN2 62.22 48.76 54.11 37.72 45.52
Our full model 62.73 49.54 54.75 37.61 45.77
Experiments: quantitative results
Model Text T/F Text MC Text All Diagram All
IGMN (Li et al., 2018) 57.41 40.00 46.88 36.35 41.36
Our full model w/o visual context 62.32 49.15 54.35 36.61 45.06
w/o SSOC(VAL) 60.82 49.08 53.72 36.53 44.72
w/o SSOC(TR+VAL) 60.72 46.34 52.02 36.57 43.93
w/o f-GCN & SSOC(TR+VAL) 58.62 44.77 50.24 35.20 42.36
Our full model w/ f-GCN2 62.22 48.76 54.11 37.72 45.52
w/o SSOC(VAL) 62.63 48.43 54.03 37.32 45.28
w/o SSOC(TR+VAL) 61.42 46.67 52.49 36.71 44.22
w/o f-GCN & SSOC(TR+VAL) 58.72 45.16 50.51 35.67 42.74
Our full model 62.73 49.54 54.75 37.61 45.77
w/o SSOC(VAL) 62.22 48.82 54.11 37.47 45.39
w/o SSOC(TR+VAL) 60.02 46.86 52.06 36.61 43.97
w/o f-GCN & SSOC(TR+VAL) 58.72 45.16 50.51 35.67 42.74
SSOC : Self-Supervised Open-set Comprehension
Experiments: quantitative results
Model Text Diagram All
Our full model w/o SSOC 52.06 36.61 43.97
w/o anchor flag (q) 49.29 35.78 42.21
w/o anchor flag (a) 43.24 31.50 37.09
w/o anchor flag (q & a) 42.64 31.72 36.92
Experiments: qualitative results
… lithosphere and asthenosphere are
layers based on physical properties .
the outermost layer is the lithosphere .
the lithosphere is the crust and the
uppermost mantle . in terms of physical
properties , this layer is rigid , solid ,
and brittle . it is easily cracked or
broken . below the lithosphere is the
asthenosphere . the asthenosphere is
also in the upper mantle . this layer is
solid , but it can flow and bend . a
solid that can flow is like silly putty ..…
Q what layer is directly below the crust ?
a) mantle
b) core
c) inner layer
d) space
asthenosphere
crust
…
…
uppermost
lithosphere
mantle
below
layer
outer
middle
layer
mantle
crust
Experiments: qualitative results
… runoff carved channels in the soil in
figure 19.1 . running water causes most
soil erosion , but wind can carry soil
away too . what humans do to soil
makes it more or less likely to be eroded
by wind or water . human actions that
can increase soil erosion are described
below .…
Q the main cause of soil erosion is ____
a) wind .
b) ice wedging .
c) abrasion .
d) running water .
causes
water
running
…
…
erosion
soil
Experiments: qualitative results
… the dense , iron core forms the center
of the earth . scientists know that the
core is metal from studying metallic
meteorites and the earths density .
seismic waves show that the outer core
is liquid , while the inner core is solid .
movement within earths outer liquid iron
core creates earths magnetic field . these
convection currents form in the outer
core because the base of the outer core
is heated by the even hotter inner core…
Q convection currents occur in the inner core .
a) true
b) false
form
core
currents
outer convection
these
in
……
Conclusion & Further work
Multi-modal understanding for TQA problems
more paragraphs
visual information
Self-supervised open-set comprehension
without Q-A-Context triple
State-of-the-art performance
still a lot of room for improvement
Textbook Question Answering (TQA) with Multi-modal Context Graph Understanding and Self-supervised Open-set Comprehension

More Related Content

Similar to Textbook Question Answering (TQA) with Multi-modal Context Graph Understanding and Self-supervised Open-set Comprehension

Step-by-step approach to question answering
Step-by-step approach to question answeringStep-by-step approach to question answering
Step-by-step approach to question answering
NAVER Engineering
 
A navigation mesh_for_dynamic_environments
A navigation mesh_for_dynamic_environmentsA navigation mesh_for_dynamic_environments
A navigation mesh_for_dynamic_environments
Roger Hernando Buch
 
Aerospace Engineering (AE) - Gate Previous Question Paper 2011 Download
Aerospace Engineering (AE) - Gate Previous Question Paper 2011 DownloadAerospace Engineering (AE) - Gate Previous Question Paper 2011 Download
Aerospace Engineering (AE) - Gate Previous Question Paper 2011 Download
Rakesh Bhupathi
 
Civil engineering mock test
Civil engineering mock testCivil engineering mock test
Civil engineering mock test
akshay015
 
Graph Modification: Beyond the known Boundaries
Graph Modification: Beyond the known BoundariesGraph Modification: Beyond the known Boundaries
Graph Modification: Beyond the known Boundaries
AkankshaAgrawal55
 
Cs 2003
Cs 2003Cs 2003
Cs 2003
Ravi Rajput
 
Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)
Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)
Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)
Thoma Itoh
 
Presentation IDETC
Presentation IDETCPresentation IDETC
Presentation IDETCXiao Wang
 
Ugc net solutions at target ies
Ugc net solutions at target iesUgc net solutions at target ies
Ugc net solutions at target ies
neeraj7svp
 
Deep learning for molecules, introduction to chainer chemistry
Deep learning for molecules, introduction to chainer chemistryDeep learning for molecules, introduction to chainer chemistry
Deep learning for molecules, introduction to chainer chemistry
Kenta Oono
 
Nift ug technology sample question paper gat
Nift ug technology sample question paper gatNift ug technology sample question paper gat
Nift ug technology sample question paper gat
Academy of Fashion & Design
 
Midterm_2013_formatted.pdf
Midterm_2013_formatted.pdfMidterm_2013_formatted.pdf
Midterm_2013_formatted.pdf
HassanJOudah
 
(Www.entrance exam.net)-gate civil engineering sample paper 4
(Www.entrance exam.net)-gate civil engineering sample paper 4(Www.entrance exam.net)-gate civil engineering sample paper 4
(Www.entrance exam.net)-gate civil engineering sample paper 4gulaaab
 
Predicting Subsystem Defects using Dependency Graph Complexities
Predicting Subsystem Defects using Dependency Graph Complexities Predicting Subsystem Defects using Dependency Graph Complexities
Predicting Subsystem Defects using Dependency Graph Complexities
Thomas Zimmermann
 
Split Contraction: The Untold Story
Split Contraction: The Untold StorySplit Contraction: The Untold Story
Split Contraction: The Untold Story
AkankshaAgrawal55
 
Lecture 2
Lecture 2Lecture 2
Lecture 2
chandsek666
 
Reasoning Questions for SSC and other Competitive Aspirants
Reasoning Questions for SSC and other Competitive AspirantsReasoning Questions for SSC and other Competitive Aspirants
Reasoning Questions for SSC and other Competitive Aspirants
DeepChand32
 
3rd Semester Computer Science and Engineering (ACU) Question papers
3rd Semester Computer Science and Engineering  (ACU) Question papers3rd Semester Computer Science and Engineering  (ACU) Question papers
3rd Semester Computer Science and Engineering (ACU) Question papers
BGS Institute of Technology, Adichunchanagiri University (ACU)
 
Pgcet Computer Science 2015 question paper
Pgcet Computer Science 2015 question paperPgcet Computer Science 2015 question paper
Pgcet Computer Science 2015 question paper
Eneutron
 

Similar to Textbook Question Answering (TQA) with Multi-modal Context Graph Understanding and Self-supervised Open-set Comprehension (20)

Step-by-step approach to question answering
Step-by-step approach to question answeringStep-by-step approach to question answering
Step-by-step approach to question answering
 
A navigation mesh_for_dynamic_environments
A navigation mesh_for_dynamic_environmentsA navigation mesh_for_dynamic_environments
A navigation mesh_for_dynamic_environments
 
Aerospace Engineering (AE) - Gate Previous Question Paper 2011 Download
Aerospace Engineering (AE) - Gate Previous Question Paper 2011 DownloadAerospace Engineering (AE) - Gate Previous Question Paper 2011 Download
Aerospace Engineering (AE) - Gate Previous Question Paper 2011 Download
 
Civil engineering mock test
Civil engineering mock testCivil engineering mock test
Civil engineering mock test
 
Graph Modification: Beyond the known Boundaries
Graph Modification: Beyond the known BoundariesGraph Modification: Beyond the known Boundaries
Graph Modification: Beyond the known Boundaries
 
Cs 2003
Cs 2003Cs 2003
Cs 2003
 
Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)
Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)
Friedlander et al. Evolution of Bow-Tie Architectures in Biology (2015)
 
Presentation IDETC
Presentation IDETCPresentation IDETC
Presentation IDETC
 
Ugc net solutions at target ies
Ugc net solutions at target iesUgc net solutions at target ies
Ugc net solutions at target ies
 
Deep learning for molecules, introduction to chainer chemistry
Deep learning for molecules, introduction to chainer chemistryDeep learning for molecules, introduction to chainer chemistry
Deep learning for molecules, introduction to chainer chemistry
 
Nift ug technology sample question paper gat
Nift ug technology sample question paper gatNift ug technology sample question paper gat
Nift ug technology sample question paper gat
 
Midterm_2013_formatted.pdf
Midterm_2013_formatted.pdfMidterm_2013_formatted.pdf
Midterm_2013_formatted.pdf
 
(Www.entrance exam.net)-gate civil engineering sample paper 4
(Www.entrance exam.net)-gate civil engineering sample paper 4(Www.entrance exam.net)-gate civil engineering sample paper 4
(Www.entrance exam.net)-gate civil engineering sample paper 4
 
Predicting Subsystem Defects using Dependency Graph Complexities
Predicting Subsystem Defects using Dependency Graph Complexities Predicting Subsystem Defects using Dependency Graph Complexities
Predicting Subsystem Defects using Dependency Graph Complexities
 
4th semester Civil Engineering (2013-June) Question Papers
4th semester Civil Engineering (2013-June) Question Papers 4th semester Civil Engineering (2013-June) Question Papers
4th semester Civil Engineering (2013-June) Question Papers
 
Split Contraction: The Untold Story
Split Contraction: The Untold StorySplit Contraction: The Untold Story
Split Contraction: The Untold Story
 
Lecture 2
Lecture 2Lecture 2
Lecture 2
 
Reasoning Questions for SSC and other Competitive Aspirants
Reasoning Questions for SSC and other Competitive AspirantsReasoning Questions for SSC and other Competitive Aspirants
Reasoning Questions for SSC and other Competitive Aspirants
 
3rd Semester Computer Science and Engineering (ACU) Question papers
3rd Semester Computer Science and Engineering  (ACU) Question papers3rd Semester Computer Science and Engineering  (ACU) Question papers
3rd Semester Computer Science and Engineering (ACU) Question papers
 
Pgcet Computer Science 2015 question paper
Pgcet Computer Science 2015 question paperPgcet Computer Science 2015 question paper
Pgcet Computer Science 2015 question paper
 

More from LGCNSairesearch

KorQuAD v2.0 소개
KorQuAD v2.0 소개KorQuAD v2.0 소개
KorQuAD v2.0 소개
LGCNSairesearch
 
Episodic Memory Reader: Learning What to Remember for Question Answering from...
Episodic Memory Reader: Learning What to Remember for Question Answering from...Episodic Memory Reader: Learning What to Remember for Question Answering from...
Episodic Memory Reader: Learning What to Remember for Question Answering from...
LGCNSairesearch
 
[saltlux] KorQuAD v1.0 참관기
[saltlux] KorQuAD v1.0 참관기[saltlux] KorQuAD v1.0 참관기
[saltlux] KorQuAD v1.0 참관기
LGCNSairesearch
 
KorQuAD v1.0 Turn up
KorQuAD v1.0 Turn upKorQuAD v1.0 Turn up
KorQuAD v1.0 Turn up
LGCNSairesearch
 
On-Device AI
On-Device AIOn-Device AI
On-Device AI
LGCNSairesearch
 
NLU Tech Talk with KorBERT
NLU Tech Talk with KorBERTNLU Tech Talk with KorBERT
NLU Tech Talk with KorBERT
LGCNSairesearch
 
딥러닝 기반의 자연어처리 최근 연구 동향
딥러닝 기반의 자연어처리 최근 연구 동향딥러닝 기반의 자연어처리 최근 연구 동향
딥러닝 기반의 자연어처리 최근 연구 동향
LGCNSairesearch
 

More from LGCNSairesearch (7)

KorQuAD v2.0 소개
KorQuAD v2.0 소개KorQuAD v2.0 소개
KorQuAD v2.0 소개
 
Episodic Memory Reader: Learning What to Remember for Question Answering from...
Episodic Memory Reader: Learning What to Remember for Question Answering from...Episodic Memory Reader: Learning What to Remember for Question Answering from...
Episodic Memory Reader: Learning What to Remember for Question Answering from...
 
[saltlux] KorQuAD v1.0 참관기
[saltlux] KorQuAD v1.0 참관기[saltlux] KorQuAD v1.0 참관기
[saltlux] KorQuAD v1.0 참관기
 
KorQuAD v1.0 Turn up
KorQuAD v1.0 Turn upKorQuAD v1.0 Turn up
KorQuAD v1.0 Turn up
 
On-Device AI
On-Device AIOn-Device AI
On-Device AI
 
NLU Tech Talk with KorBERT
NLU Tech Talk with KorBERTNLU Tech Talk with KorBERT
NLU Tech Talk with KorBERT
 
딥러닝 기반의 자연어처리 최근 연구 동향
딥러닝 기반의 자연어처리 최근 연구 동향딥러닝 기반의 자연어처리 최근 연구 동향
딥러닝 기반의 자연어처리 최근 연구 동향
 

Recently uploaded

原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样
原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样
原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样
u86oixdj
 
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
u86oixdj
 
Best best suvichar in gujarati english meaning of this sentence as Silk road ...
Best best suvichar in gujarati english meaning of this sentence as Silk road ...Best best suvichar in gujarati english meaning of this sentence as Silk road ...
Best best suvichar in gujarati english meaning of this sentence as Silk road ...
AbhimanyuSinha9
 
Learn SQL from basic queries to Advance queries
Learn SQL from basic queries to Advance queriesLearn SQL from basic queries to Advance queries
Learn SQL from basic queries to Advance queries
manishkhaire30
 
Influence of Marketing Strategy and Market Competition on Business Plan
Influence of Marketing Strategy and Market Competition on Business PlanInfluence of Marketing Strategy and Market Competition on Business Plan
Influence of Marketing Strategy and Market Competition on Business Plan
jerlynmaetalle
 
一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理
一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理
一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理
mzpolocfi
 
Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)
TravisMalana
 
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdfCh03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
haila53
 
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTESAdjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
Subhajit Sahu
 
Analysis insight about a Flyball dog competition team's performance
Analysis insight about a Flyball dog competition team's performanceAnalysis insight about a Flyball dog competition team's performance
Analysis insight about a Flyball dog competition team's performance
roli9797
 
哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样
哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样
哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样
axoqas
 
Unleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdf
Unleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdfUnleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdf
Unleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdf
Enterprise Wired
 
My burning issue is homelessness K.C.M.O.
My burning issue is homelessness K.C.M.O.My burning issue is homelessness K.C.M.O.
My burning issue is homelessness K.C.M.O.
rwarrenll
 
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Subhajit Sahu
 
The Building Blocks of QuestDB, a Time Series Database
The Building Blocks of QuestDB, a Time Series DatabaseThe Building Blocks of QuestDB, a Time Series Database
The Building Blocks of QuestDB, a Time Series Database
javier ramirez
 
一比一原版(CBU毕业证)卡普顿大学毕业证如何办理
一比一原版(CBU毕业证)卡普顿大学毕业证如何办理一比一原版(CBU毕业证)卡普顿大学毕业证如何办理
一比一原版(CBU毕业证)卡普顿大学毕业证如何办理
ahzuo
 
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
dwreak4tg
 
【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】
【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】
【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】
NABLAS株式会社
 
Machine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptxMachine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptx
balafet
 
Nanandann Nilekani's ppt On India's .pdf
Nanandann Nilekani's ppt On India's .pdfNanandann Nilekani's ppt On India's .pdf
Nanandann Nilekani's ppt On India's .pdf
eddie19851
 

Recently uploaded (20)

原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样
原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样
原版制作(swinburne毕业证书)斯威本科技大学毕业证毕业完成信一模一样
 
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
 
Best best suvichar in gujarati english meaning of this sentence as Silk road ...
Best best suvichar in gujarati english meaning of this sentence as Silk road ...Best best suvichar in gujarati english meaning of this sentence as Silk road ...
Best best suvichar in gujarati english meaning of this sentence as Silk road ...
 
Learn SQL from basic queries to Advance queries
Learn SQL from basic queries to Advance queriesLearn SQL from basic queries to Advance queries
Learn SQL from basic queries to Advance queries
 
Influence of Marketing Strategy and Market Competition on Business Plan
Influence of Marketing Strategy and Market Competition on Business PlanInfluence of Marketing Strategy and Market Competition on Business Plan
Influence of Marketing Strategy and Market Competition on Business Plan
 
一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理
一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理
一比一原版(Dalhousie毕业证书)达尔豪斯大学毕业证如何办理
 
Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)
 
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdfCh03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
 
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTESAdjusting OpenMP PageRank : SHORT REPORT / NOTES
Adjusting OpenMP PageRank : SHORT REPORT / NOTES
 
Analysis insight about a Flyball dog competition team's performance
Analysis insight about a Flyball dog competition team's performanceAnalysis insight about a Flyball dog competition team's performance
Analysis insight about a Flyball dog competition team's performance
 
哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样
哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样
哪里卖(usq毕业证书)南昆士兰大学毕业证研究生文凭证书托福证书原版一模一样
 
Unleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdf
Unleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdfUnleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdf
Unleashing the Power of Data_ Choosing a Trusted Analytics Platform.pdf
 
My burning issue is homelessness K.C.M.O.
My burning issue is homelessness K.C.M.O.My burning issue is homelessness K.C.M.O.
My burning issue is homelessness K.C.M.O.
 
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
Algorithmic optimizations for Dynamic Levelwise PageRank (from STICD) : SHORT...
 
The Building Blocks of QuestDB, a Time Series Database
The Building Blocks of QuestDB, a Time Series DatabaseThe Building Blocks of QuestDB, a Time Series Database
The Building Blocks of QuestDB, a Time Series Database
 
一比一原版(CBU毕业证)卡普顿大学毕业证如何办理
一比一原版(CBU毕业证)卡普顿大学毕业证如何办理一比一原版(CBU毕业证)卡普顿大学毕业证如何办理
一比一原版(CBU毕业证)卡普顿大学毕业证如何办理
 
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
 
【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】
【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】
【社内勉強会資料_Octo: An Open-Source Generalist Robot Policy】
 
Machine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptxMachine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptx
 
Nanandann Nilekani's ppt On India's .pdf
Nanandann Nilekani's ppt On India's .pdfNanandann Nilekani's ppt On India's .pdf
Nanandann Nilekani's ppt On India's .pdf
 

Textbook Question Answering (TQA) with Multi-modal Context Graph Understanding and Self-supervised Open-set Comprehension

  • 1. Daesik Kim*, Seonhoon Kim*, Nojun Kwak
  • 3. Textbook QA? The most realistic QA problem !
  • 4. Textbook QA? The most complicated QA problem !
  • 6. … and the title is … Q What is the title of this paper? https://media-assets.bookbub.com/wp-content/uploads/ 2015/12/cute-baby-skim-reading-gif.gif Context Solving step
  • 8. Q Plates move over Earth’s surface because of _________ a) conduction within the crust. b) radiation from the inner core. c) subduction in the outer core. d) convection within the mantle. http://www.grygla.k12.mn.us/uploads/2/3/7/1/23718473/earth_sci_-_ch._6_-_plate_tectonics.pdf Context
  • 9. Q Plates move over Earth’s surface because of _________ a) conduction within the crust. b) radiation from the inner core. c) subduction in the outer core. d) convection within the mantle. Q What is the southern most point of the ring of fire? a) South sandwich trench b) Japan trench c) Aleutian trench d) Kurile trench http://www.grygla.k12.mn.us/uploads/2/3/7/1/23718473/earth_sci_-_ch._6_-_plate_tectonics.pdf
  • 10. (Kembhavi et al., 2017) single sentence / multiple sentences within a paragraph question/context diagram, and one paragraph
  • 13. Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 … Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4
  • 14. Q & a#1 & C#2 Correct? Wrong? Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4 Top-1 Tf-idf score … model
  • 15. Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4 Top-1 Tf-idf score … Q & a#1 & C#2 Correct? Wrong? Q & a#2 & C#1 Correct? Wrong? model
  • 16. Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4 Top-1 Tf-idf score … Q & a#1 & C#2 Correct? Wrong? Q & a#2 & C#1 Correct? Wrong? Q & a#3 & C#2 Correct? Wrong? model
  • 17. Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4 Top-1 Tf-idf score … Q & a#1 & C#2 Correct? Wrong? Q & a#2 & C#1 Correct? Wrong? Q & a#3 & C#2 Correct? Wrong? Q & a#4 & C#9 Correct? Wrong? model
  • 18. Self-supervised comprehension Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 … Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4 Top-1 Tf-idf score Q & a#1 → C#2 (Top-1 tf-idf score) → C#4 (Top-2 tf-idf score) → C#1 (Top-3 tf-idf score) → C#7 (Top-4 tf-idf score) → C#5 (Top-5 tf-idf score)
  • 19. Q & a#1 & C#2 Correct Q & a#1 & C#4 Wrong Q & a#1 & C#1 Wrong Q & a#1 & C#7 Wrong Q & a#1 & C#5 Wrong Self-supervised comprehension Context Paragraph #1 Paragraph #2 Paragraph #3 Paragraph #9 Paragraph #10 Q What is the question? a) answer option #1 b) answer option #2 c) answer option #3 d) answer option #4 Q & a#1 & C#2 Correct? Wrong? Q & a#1 → C#2 (Top-1 tf-idf score) → C#4 (Top-2 tf-idf score) → C#1 (Top-3 tf-idf score) → C#7 (Top-4 tf-idf score) → C#5 (Top-5 tf-idf score) Top-1 Tf-idf score… m model
  • 21. Multi-modal context graph understanding: visual context rabbit connects to fox mouse ladybird snake 14 objects 22 stages … diagram parsing from UDPnet (Kim et al., 2018) parsed info from UDPnet & OCR info from TQA build context graph of diagrams rabbit fox mouseladybird 14 objects 22 stages context matrix 𝑪 𝒅 adjacency matrix 𝑨 𝒅
  • 22. Multi-modal context graph understanding: textual context … and the title is … Context Q What is the title of this paper? anchor node dependency parsing filtering by anchor nodes context matrix 𝑪 𝒕 adjacency matrix 𝑨𝒕 3 1 1 1 1 2 2
  • 23. Multi-modal context graph understanding Question Answer word representation (word embedding; char representation; anchor flag) BiLSTM Q repr. : 𝒉 𝒒 A repr. : 𝒉 𝒂 visual Context graph textual Context graph GCN GCN GCN attention fused representation (f-GCN1) fused representation (f-GCN2) … max pooling 𝑯 𝒄 𝟏 𝑯 𝒄 𝟐 𝑨 𝒅 𝑨 𝒕 𝑨 𝒕 𝑨 𝒅 , 𝑨 𝒕 : adjacency matrix Question & Answer & Context Correct? Wrong?model
  • 24. Multi-modal problem solving 𝑘𝑡ℎ input for the solving RNN 𝒄𝒐𝒏𝒄𝒂𝒕(𝒉 𝒒; 𝒉 𝒂; 𝑨𝒕𝒕 𝒒; 𝑨𝒕𝒕 𝒂; 𝑨𝒕𝒕 𝒒 𝒒𝒅 ; 𝑨𝒕𝒕 𝒂 𝒒𝒅 ) BiLSTM FC ෝ𝒚 𝟏 ෝ𝒚 𝒌… 𝒌 𝒕𝒉 𝑨𝒕𝒕 𝒒 = 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒒, 𝑯 𝒄) 𝑨𝒕𝒕 𝒂 = 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒂, 𝑯 𝒄) 𝑨𝒕𝒕 𝒒 𝒒𝒅 = 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒒, 𝑯 𝒒𝒅) 𝑨𝒕𝒕 𝒂 𝒒𝒅 = 𝒂𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏(𝒉 𝒂, 𝑯 𝒒𝒅) only for the diagram question
  • 26. Experiments: quantitative results Model Text T/F Text MC Text All Diagram All Random 50.10 22.88 33.62 24.96 29.08 MemN+VQA (Kembhavi et al., 2017) 50.50 31.05 38.73 31.82 35.11 MemN+DPG (Kembhavi et al., 2017) 50.50 30.98 38.69 32.83 35.62 BiDAF+DPG (Kembhavi et al., 2017) 50.40 30.46 38.33 32.72 35.39 Challenge - - 45.57 35.85 40.48 IGMN (Li et al., 2018) 57.41 40.00 46.88 36.35 41.36 Our full model w/o visual context 62.32 49.15 54.35 36.61 45.06 Our full model w/ f-GCN2 62.22 48.76 54.11 37.72 45.52 Our full model 62.73 49.54 54.75 37.61 45.77
  • 27. Experiments: quantitative results Model Text T/F Text MC Text All Diagram All IGMN (Li et al., 2018) 57.41 40.00 46.88 36.35 41.36 Our full model w/o visual context 62.32 49.15 54.35 36.61 45.06 w/o SSOC(VAL) 60.82 49.08 53.72 36.53 44.72 w/o SSOC(TR+VAL) 60.72 46.34 52.02 36.57 43.93 w/o f-GCN & SSOC(TR+VAL) 58.62 44.77 50.24 35.20 42.36 Our full model w/ f-GCN2 62.22 48.76 54.11 37.72 45.52 w/o SSOC(VAL) 62.63 48.43 54.03 37.32 45.28 w/o SSOC(TR+VAL) 61.42 46.67 52.49 36.71 44.22 w/o f-GCN & SSOC(TR+VAL) 58.72 45.16 50.51 35.67 42.74 Our full model 62.73 49.54 54.75 37.61 45.77 w/o SSOC(VAL) 62.22 48.82 54.11 37.47 45.39 w/o SSOC(TR+VAL) 60.02 46.86 52.06 36.61 43.97 w/o f-GCN & SSOC(TR+VAL) 58.72 45.16 50.51 35.67 42.74 SSOC : Self-Supervised Open-set Comprehension
  • 28. Experiments: quantitative results Model Text Diagram All Our full model w/o SSOC 52.06 36.61 43.97 w/o anchor flag (q) 49.29 35.78 42.21 w/o anchor flag (a) 43.24 31.50 37.09 w/o anchor flag (q & a) 42.64 31.72 36.92
  • 29. Experiments: qualitative results … lithosphere and asthenosphere are layers based on physical properties . the outermost layer is the lithosphere . the lithosphere is the crust and the uppermost mantle . in terms of physical properties , this layer is rigid , solid , and brittle . it is easily cracked or broken . below the lithosphere is the asthenosphere . the asthenosphere is also in the upper mantle . this layer is solid , but it can flow and bend . a solid that can flow is like silly putty ..… Q what layer is directly below the crust ? a) mantle b) core c) inner layer d) space asthenosphere crust … … uppermost lithosphere mantle below layer outer middle layer mantle crust
  • 30. Experiments: qualitative results … runoff carved channels in the soil in figure 19.1 . running water causes most soil erosion , but wind can carry soil away too . what humans do to soil makes it more or less likely to be eroded by wind or water . human actions that can increase soil erosion are described below .… Q the main cause of soil erosion is ____ a) wind . b) ice wedging . c) abrasion . d) running water . causes water running … … erosion soil
  • 31. Experiments: qualitative results … the dense , iron core forms the center of the earth . scientists know that the core is metal from studying metallic meteorites and the earths density . seismic waves show that the outer core is liquid , while the inner core is solid . movement within earths outer liquid iron core creates earths magnetic field . these convection currents form in the outer core because the base of the outer core is heated by the even hotter inner core… Q convection currents occur in the inner core . a) true b) false form core currents outer convection these in ……
  • 32. Conclusion & Further work Multi-modal understanding for TQA problems more paragraphs visual information Self-supervised open-set comprehension without Q-A-Context triple State-of-the-art performance still a lot of room for improvement