Learning to Ask Good Questions: Ranking Clarification Questions using Neural Expected Expect ed Value of Perfect Perf ect Info Information rmation Sudha Rao1 and Hal Daumé III1,2 1University
of Maryland
Colle leg ge Pa Park 2Microsoft
Research
New Yor York k
CODE + DATA: https://github.com/raosudha89/ranking_clarification_questions
Natural Language Understanding
2
Natural Language Understanding
How long does it take to get a PhD?
3
Natural Language Understanding Give me a recipe for lasagna
How long does it take to get a PhD?
4
Natural Language Understanding Give me a recipe for lasagna
How long does it take to get a PhD?
Please bring me my coffee mug from f rom the kitchen
5
Natural Language Understanding Give me a recipe for lasagna
How long does it take to get a PhD?
Please bring me my coffee mug from f rom the kitchen
6
Human Interactions
7
Human Interactions
Please bring me my coffee mug from the kitchen
8
Human Interactions
Please bring me my coffee mug from the kitchen
9
Human Interactions
Please bring me my coffee mug from the kitchen
What color is your your coffee coff ee mug?
10
Teach Machines to Ask Clarification Clarification Questions
11
Teach Machines to Ask Clarification Questions Questions
How long does it take to get a PhD ? In which field?
12
Teach Machines to Ask Clarification Questions Questions
How long does it take to get a PhD ? In which field?
Give me a recipe for lasagna Any dietary restrictions?
13
Teach Machines to Ask Clarification Questions Questions
How long does it take to get a PhD ? In which field?
Please bring me my coffee mug from the kitchen
Give me a recipe for lasagna Any dietary restrictions?
What color is your coffee coff ee mug?
14
Teach Machines to Ask Clarification Questions Questions Context-aware questions about missing information
How long does it take to get a PhD ? In which field?
Please bring me my coffee mug from the kitchen
Give me a recipe for lasagna Any dietary restrictions?
What color is your coffee coff ee mug?
15
Reading Comprehension Question Generation My class is going to the movies on a field trip tr ip next week. We have to get permission slips signed before we go. We are going to see a movie that tells the story from a
someone’s Goal: Assess someone’s understanding of the text
book we read.
Q: What do the students need to do before before going to the movies? movies?
o
o
Heilman. “Automatic factual question generation from text” Ph.D. thesis 2011 Vasile, Vasile, et al. "The first question question generation generation shar shared ed task evaluation evaluation challenge. challenge.” ” NLG 2010
o
Olney, Graesser, and Person. "Question generation from concept maps." Dialogue & Discourse 2012
o
Chali and Hasan. "Towards Topic-to-Question Generation." ACL 2015
o
Serban, et al. "Generating Factoid Questions With Recurrent Recur rent Neural Networks” ACL 2016
o
Du, Shao & Cardie "Learning to ask: Neural question generation for reading comprehension" ACL 2017
o
Tang et al. "Learning "Learni ng to Collaborate for Question Answering and Asking." NAACL 2018
o
Mrinmaya and Xing. "Self-Training for Jointly Learning to Ask and Answer Questions." NAACL 2018
16
Question Generation Generation for Slot Filling SLOTS USER:
I want to go to Melbourne on July 14
SYSTEM:
What time do you want to leave?
USER:
I must be in Melbourne rne by 11 am
SYSTEM SYS TEM::
Would ould you like like a Delta flight that arrive arr ives s at 10.15 am?
USER:
Sure
SYS YST TEM: EM:
In what name should I make the reservation?
o
Goddeau, et al. "A form-based dialogue manager for spoken language applications.” 1996
o
Bobrow., et al. "GUS, a frame-driven dialog system." Artificial intelligence 1977
o
Lemon, et al. "An ISU dialogue system exhibiting reinforcement learning of dialogue policies: generic slot-filling in the TALK in-car system.” EACL 2006
o
o
o
Williams, et al. The Dialog State Tracking Challenge” SIGDIAL 2013 Young Young,, et al. “Pomdp-based “Pomdp-based statistical statistical spoken spoken dialog systems: systems: A review review..” IEEE 2013 Dhingra, et al. "Towards End-to-End Reinforcement Learning of Dialogue Agents for Information Access.” Access.” ACL 2017
o
Bordes, et al. "Learning "Learni ng end-to-end goal-oriented goal-or iented dialog.” dialog.” ICLR 2017
17
Other types of Question Generation o
Liu, et al. “Automatic question generation for literature review writing support." International Conference on Intelligent Tutoring Systems. 2010
o
Penas and Hovy, “Filling knowledge gaps in text for machine reading” International Conference on Computational Linguistics: Posters ACL 2010
o
Artzi & Zettlemoyer Zettlemoyer,, “Bootstrapping semantic semantic parsers from conversations” conversations” EMNLP 2011
o
Labutov, Labutov, et al.“Deep questions question s without deep understanding” understa nding” ACL 2015
o
Mostafazadeh et al. "Generating "Generating natural questions about an image." i mage." ACL 2016
o
Mostafazadeh et al. a l. "Multimodal Context for Natural Nat ural Question and Response Generation.” Generation.” IJCNLP 2017. 2017 .
o
Rothe, Lake and Gureckis. “Question asking as program generation” NIPS 2017.
18
Talk Outline o
Clarification Questions Dataset
o
Problem Formulation: ormulatio n: Question Ranking
o
Expected Value of Perfect Information (EVPI) inspired model
o
Evaluation
o
Conclusion
19
Talk Outline o
Clarification Questions Dataset
o
Problem Formulation: ormulatio n: Question Ranking
o
Expected Value of Perfect Information (EVPI) inspired model
o
Evaluation
o
Conclusion
20
Clarification Questions Dataset
StackExchange Question-Answer Forum
21
Clarification Questions Dataset
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
22
Clarification Questions Dataset
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
Finding: Questions go unanswered for a long time if they are not clear enough Asaduzzaman, Asaduzzaman, Muhammad, et al. "Answering "Answering questions about about unanswer unanswered ed questions of stack overflo overflow w.” Working Conference on Mining Software Repositories. IEEE Press, 2013.
23
Clarification Questions Dataset
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
Question comment
What version of ubuntu do you have?
24
Clarification Questions Dataset
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation.
Initial Post
I'm having this error message while running ./configure
Question comment
So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance! What version of ubuntu do you have? I'm aiming to install ape in Ubuntu 14.04 LTS, a simple code for pseudopotential generation. I'm having this error message while running ./configure
Updated Post
So I have the library but the program program installation isn't finding it. Any help? Thanks in advance!
25
Clarification Questions Dataset
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation.
Initial Post
I'm having this error message while running ./configure
Question comment
So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance! What version of ubuntu do you have? I'm aiming to install ape in Ubuntu 14.04 LTS, a simple code for pseudopotential generation.
Edit as an answer to the question
I'm having this error message while running ./configure
Updated Post
So I have the library but the program program installation isn't finding it. Any help? Thanks in advance!
26
Clarification Questions Dataset
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation.
Initial Post
I'm having this error message while running ./configure
Question comment
So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance! What version of ubuntu do you have? I'm aiming to install ape in Ubuntu 14.04 LTS, a simple code for pseudopotential generation.
Edit as an answer to the question
I'm having this error message while running ./configure
Updated Post
So I have the library but the program program installation isn't finding it. Any help? Thanks in advance!
27
Clarification Questions Dataset Dataset Creation
(
post
,
question
,
answer
) triples
post
Original post
question
Clarification question posted in comments
answer
Edit made to the post in response response to the question OR author’s author’s rreply eply to the question comment
Dataset Size: ~77 K triples trip les Domains:
Askubuntu, Unix, Superuser
Note: We identify a question questio n using the question mark (?) token. We match the edit to the answer using timestamp & word word embedding similarity based heuristics. 28
Talk Outline o
Clarification Questions Dataset
o
Problem Formulation: ormulatio n: Question Ranking
o
Expected Value of Perfect Information (EVPI) inspired model
o
Evaluation
o
Conclusion
29
Problem Formulati ormulation: on: Question Question Ranking Ranking
Post
How to configure path or set environment variables? …
30
Problem Formulati ormulation: on: Question Question Ranking Ranking Generate Question Candidates Post
How to configure path or set environment variables? …
What is the make of your wifi card?
What version of Ubuntu do you have?
What OS are you using?
31
Problem Formulati ormulation: on: Question Question Ranking Ranking Generate Question Candidates Post
How to configure path or set environment variables? …
What is the make of your wifi card?
What version of Ubuntu do you have?
What OS are you using?
Rank the question candidates What version of Ubuntu do you have?
What OS are you using?
What is the make of your wifi card?
32
Problem Formulati ormulation: on: Question Question Ranking Ranking
How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
What version of Ubuntu do you have? How are you installing ape?
Shortlist of useful questions
Do you have GSL installed?
33
Talk Outline o
Clarification Questions Dataset
o
Problem Formulation: ormulatio n: Question Ranking
o
Expected Value of Perfect Information (EVPI) inspired model
o
Evaluation
o
Conclusion
34
Expected Value of Perfect Information (EVPI) inspired model
Key Idea How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
Possible questions (a) What version of Ubuntu do you have?
!
Just right
35
Expected Value of Perfect Information (EVPI) inspired model
Key Idea How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
Possible questions (a) What hat vers versio ion n of Ubun Ubuntu tu do you have? ve?
!
(b) What is the make of your wifi card?
!
Just right Not useful
36
Expected Value of Perfect Information (EVPI) inspired model
Key Idea How to configure path or set environment variables for installation? I'm aiming to install ape, a simple code for f or pseudopotential generation. I'm having this error message while running ./configure So I have the library but the program installation isn't finding it. Any help? help? Thanks in advance! advance!
Possible questions (a) What hat vers versio ion n of Ubun Ubuntu tu do you have? ve?
!
(b) What is the make of your wifi card?
!
(c) Are you running Ubuntu Ubuntu 14.10 kernel 4.4.0 4.4.0-59-59! generic on an x86 64 architecture?
Just right Not useful
Unlikely to add value
37
Expected Value of Perfect Information (EVPI) inspired model
o
Use EVPI to identify questions that add the most value to the given post
Avriel, Avriel, Mordecai, Mordecai, and A. C. Williams. "The value of information and stochastic programming." programming." Operations Research 18.5 (1970)
38
Expected Value of Perfect Information (EVPI) inspired model
o
Use EVPI to identify questions that add the most value to the given post
o
Definition: Definit ion: Value of Perfect Information Infor mation VPI (x) How How much value value does x add add to a given information content content c?
Avriel, Avriel, Mordecai, Mordecai, and A. C. Williams. "The value of information and stochastic programming." programming." Operations Research 18.5 (1970)
39
Expected Value of Perfect Information (EVPI) inspired model
o
Use EVPI to identify questions that add the most value to the given post
o
Definition: Definit ion: Value of Perfect Information Infor mation VPI (x) How How much value value does x add add to a given information content content c?
o
Since we have not acquired x, we define its value in expectation
Avriel, Avriel, Mordecai, Mordecai, and A. C. Williams. "The value of information and stochastic programming." programming." Operations Research 18.5 (1970)
40
Expected Value of Perfect Information (EVPI) inspired model
o
Use EVPI to identify questions that add the most value to the given post
o
Definition: Definit ion: Value of Perfect Information Infor mation VPI (x) How How much value value does x add add to a given information content content c?
o
Since we have not acquired x, we define its value in expectation Likelihood of x given c
EVPI (x|c) =
P (x|c) Utility(x, Utility (x, c) x
X Value Value of updating c with x
Avriel, Avriel, Mordecai, Mordecai, and A. C. Williams. "The value of information and stochastic programming." programming." Operations Research 18.5 (1970)
41
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
42
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
EVPI ( qi | p
)
p
: given post
qi
: question from set of question candidates Q
43
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
Likelihood of a j being the answer to q i on post p
EVPI ( qi | p
)=
P(
a j |
p
, qi
)
p
: given post
qi
: question from set of question candidates Q
44
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
Likelihood of a j being the answer to q i on post p
EVPI ( qi | p
)=
P(
a j |
p
, qi
) U(
p
+
a j
)
Utility of updating the post p with answer a j
p
: given post
qi
: question from set of question candidates Q
45
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
Likelihood of a j being the answer to q i on post p
EVPI ( qi | p
)= a j
P(
a j |
p
, qi
) U(
p
+
a j
)
A Utility of updating the post p with answer a j
p
: given post
qi
: question from from set of question candidates Q
a j
: answer from set of answer candidates A
46
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem We W e rank questions based on their EVPI value Likelihood of a j being the answer to q i on post p
EVPI ( qi | p
)= a j
P(
a j |
p
, qi
) U(
p
+
a j
)
A Utility of updating the post p with answer a j
p
: given post
qi
: question from from set of question candidates Q
a j
: answer from set of answer candidates A
47
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
Likelihood of a j being the answer to q i on post p
EVPI ( qi | p
)= a j
P(
a j |
p
, qi
) U(
p
+
a j
)
A Utility of updating the post p with answer a j
p
: given post
qi
: question from from set of question candidates Q
a j
: answer from set of answer candidates A
1
48
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
2 EVPI ( qi | p
Likelihood of a j being the answer to q i on post p
)= a j
P(
a j |
p
, qi
) U(
p
+
a j
)
A Utility of updating the post p with answer a j
p
: given post
qi
: question from from set of question candidates Q
a j
: answer from set of answer candidates A
1
49
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
2 EVPI ( qi | p
)= a j
P(
Answer Modeling Modeling
a j |
p
, qi
) U(
p
+
a j
)
A Utility of updating the post p with answer a j
p
: given post
qi
: question from from set of question candidates Q
a j
: answer from set of answer candidates A
1
50
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
2 EVPI ( qi | p
)= a j
P(
Answer Modeling Modeling
a j |
p
, qi
) U(
p
+
)
A 3
Utility of updating the post p with answer a j
p
: given post
qi
: question from from set of question candidates Q
a j
a j
: answer from set of answer candidates candidat es A
1
51
Expected Value of Perfect Information (EVPI) inspired model
EVPI formulation for our problem
2 EVPI ( qi | p
)= a j
P(
Answer Modeling Modeling
a j |
p
) U(
, qi
p
+
)
A 3
Utility Calculator
p
: given post
qi
: question from from set of question candidates Q
a j
a j
: answer from set of answer candidates A
1
52
Expected Value of Perfect Information (EVPI) inspired model
1
Question & Answer Candidate Generator
53
Expected Value of Perfect Information (EVPI) inspired model
1
Question & Answer Candidate Generator
Dataset of (post, question, answer)
Post as Documents
Post p as query
Lucene Search Engine
54
Expected Value of Perfect Information (EVPI) inspired model
1
Question & Answer Candidate Generator
Dataset of (post, question, answer)
Ten posts similar to given post p p1
Post as Documents
Post p as query
Lucene Search Engine
p2
p j
p10
55
Expected Value of Perfect Information (EVPI) inspired model
1
Question & Answer Candidate Generator
Dataset of (post, question, answer)
Post as Documents
Post p as query
Lucene Search Engine
Ten posts similar to given post p
Questions paired with those posts
p1
q1
p2
q2
p j
q j
p10
q10
56
Expected Value of Perfect Information (EVPI) inspired model
1
Question & Answer Candidate Generator
Dataset of (post, question, answer)
Post as Documents
Post p as query
Lucene Search Engine
Ten posts similar to given post p
Questions paired with those posts
Answers paired Answers with those posts
p1
q1
a1
p2
q2
a2
p j
q j
a j
p10
q10
a10
57
Expected Value of Perfect Information (EVPI) inspired model
1
Question & Answer Candidate Generator
Dataset of (post, question, answer)
Post as Documents
Post p as query
Lucene Search Engine
Ten posts similar to given post p
Questions paired with those posts
Answers paired Answers with those posts
p1
q1
a1
p2
q2
a2
p j
q j
a j
p10
q10
a10
58
Expected Value of Perfect Information (EVPI) inspired model
2
EVPI ( qi | p
)= a j
Answer Modeling P(
a j |
p
, qi
) U(
p
+
a j
)
A
59
Expected Value of Perfect Information (EVPI) inspired model
2 P(
a j |
p
, qi
Answer Modeling
)
60
Expected Value of Perfect Information (EVPI) inspired model
2 P(
a j |
p
, qi
Answer Modeling
)
p
qi
Post LSTM
Question LSTM
Word embedding module
p
qi
61
Expected Value of Perfect Information (EVPI) inspired model
2 P(
a j |
p
, qi
Answer Modeling
)
Embans(
p
,
qi
)
Feedforward
p
qi
Post LSTM
Question LSTM
Word embedding module
p
qi
62
Expected Value of Perfect Information (EVPI) inspired model
2 P(
a j |
p
, qi
Answer Modeling
)
!
cosine_sim(Embans(
p
,
qi
),
Feedforward
p
qi
Post LSTM
Question LSTM
a j
)
Averag Average e
Word embedding module
p
qi
a j
63
Expected Value of Perfect Information (EVPI) inspired model
2 P(
Embans(
p
a j |
, qi
)
p
, qi
Answer Modeling
)
!
cosine_sim(Embans(
p
,
qi
),
Feedforward
a j
)
Averag Average e
Training p
qi
Post LSTM
Question LSTM
Word embedding module
p
qi
a j
64
Expected Value of Perfect Information (EVPI) inspired model
2 P(
Embans(
p
a j |
, qi
p
)
Close to true a i paired with p ai : Ubuntu 14.04 LTS qi : Which version of
, qi
Answer Modeling
)
!
cosine_sim(Embans(
p
,
qi
),
Feedforward
a j
)
Averag Average e
Training p
qi
Post LSTM
Question LSTM
Ubuntu do you have? Word embedding module
p
qi
a j
65
Expected Value of Perfect Information (EVPI) inspired model
2 P(
Embans(
p
a j |
, qi
p
)
Close to true a i paired with p ai : Ubuntu 14.04 LTS
, qi
Answer Modeling
)
!
cosine_sim(Embans(
p
,
qi
),
Feedforward
a j
)
Averag Average e
Training Close to ak paired with qk similar to true qi
qi : Which version of
p
qi
Post LSTM
Question LSTM
Ubuntu do you have? ak : Ubuntu 11.10 qK : What OS are you using?
Word embedding module
p
qi
a j
66
Expected Value of Perfect Information (EVPI) inspired model
Utility Calculator
3
EVPI ( qi | p
)= a j
P( A
a j |
p
, qi
) U(
p
+
a j
)
Expected Value of Perfect Information (EVPI) inspired model
3
Utility Calculator U( p
+
a j
)
Expected Value of Perfect Information (EVPI) inspired model
3
Utility Calculator Value V alue between between 0 and 1
U( p
+
a j
)
Feedforward
p
qi
Post LSTM
Question LSTM
a j
Answer Answer LSTM
Word embedding module
p
qi
a j
Expected Value of Perfect Information (EVPI) inspired model
3
Utility Calculator Value V alue between between 0 and 1
U( p Training
+
a j
)
Feedforward
p
qi
Post LSTM
Question LSTM
a j
Answer Answer LSTM
Word embedding module
p
qi
a j
Expected Value of Perfect Information (EVPI) inspired model
3
Utility Calculator Value V alue between between 0 and 1
U( p Training
ai : Ubuntu 14.04 LTS
a j
)
Feedforward
Label qi : Which version of Ubuntu do you have?
+
y=1 p
qi
Post LSTM
Question LSTM
a j
Answer Answer LSTM
Word embedding module
p
qi
a j
Expected Value of Perfect Information (EVPI) inspired model
3
Utility Calculator Value V alue between between 0 and 1
U( p Training
ai : Ubuntu 14.04 LTS q j : What OS are you using? a j : Ubuntu 11.10
qk : What is the make of your wifi card? ak : TP-Link TL-WDN4800 TL-WDN480 0
a j
)
Feedforward
Label qi : Which version of Ubuntu do you have?
+
y=1 p
qi
Post LSTM
Question LSTM
a j
y=0 Answer Answer LSTM
Word embedding module
y=0 p
qi
a j
Expected Value of Perfect Information (EVPI) inspired model
3
Utility Calculator Value V alue between between 0 and 1
U( p Training (Minimize binary cross-entropy)
ai : Ubuntu 14.04 LTS q j : What OS are you using? a j : Ubuntu 11.10
qk : What is the make of your wifi card? ak : TP-Link TL-WDN4800 TL-WDN480 0
a j
)
Feedforward
Label qi : Which version of Ubuntu do you have?
+
y=1 p
qi
Post LSTM
Question LSTM
a j
y=0 Answer Answer LSTM
Word embedding module
y=0 p
qi
a j
Expected Value of Perfect Information (EVPI) inspired model
1
2
3
Question & Answer Candidate Generator
Answer Modeling
Utility Calculator
74
Expected Value of Perfect Information (EVPI) inspired model
1
2
3
Question & Answer Candidate Generator
Answer Modeling
Utility Calculator
Train time behavior: For each eac h (p, q, a) in our train trai n set 1. Generate question candidates (Q) and answer candidates (A) 2. Train Answer Model and Utility Calculator using joint loss loss function : lossans (p, q, a, Q) + lossutil (y, p, q, a)
75
Expected Value of Perfect Information (EVPI) inspired model
1
2
3
Question & Answer Candidate Generator
Answer Modeling
Utility Calculator
Train time behavior: For each eac h (p, q, a) in our train trai n set 1. Generate question candidates (Q) and answer candidates (A) 2. Train Answer Model and Utility Calculator using joint loss loss function : lossans (p, q, a, Q) + lossutil (y, p, q, a)
Test time behavior: Given a post from our test set 1. Generate question candidates (Q) and answer candidates (A) 2. Calculate P(a j |p, qi) for each q 3. Calculate U(p + a j) for each a
Q using Answer Model A using Utility Calculator
4. Rank questions by EVPI (q i | p) = P(a j | p, qi) U(p + a j) a j A
76
Talk Outline o
Clarification Questions Dataset
o
Problem Formulation: ormulatio n: Question Ranking
o
Expected Value of Perfect Information (EVPI) inspired model
o
Evaluation
o
Conclusion
77
Evaluation Too much disk read/write when launching an application I have Xubuntu 13.04 on an old Dell Inspiron.
Post
When I launch an application it takes a pretty long time to be launched and I see a lot of disk read/write. read/wr ite. If the system was short on memory, this would be understandable as the system would use swap. But that's not the case in my situation (i.e. I have this problem even when the RAM is almost empty).
Question Candidates 1.
How much ram do you have installed ? and what size it the swap disk partition ?
2.
If you do not have any problem with getting a little techy then may i suggest a method ?
3.
How is it slow exactly ? boot time ? hdd read/write ? cpu time ? graphics rendering ?
4.
What is the longest time you have let it run ?
5.
This may be a silly question but ... did you make your usb stick bootable ?
6.
Do your system were recently updated ?
7.
Why not have two ssds in raid 1 for redundancy ?
8.
Is that a `parted -- list` on the synology device ?
9.
Can you tell us a little about your configuration ?
10. Did you turn hardware virtualization on in bios/efi ?
78
Evaluation Too much disk read/write when launching an application I have Xubuntu 13.04 on an old Dell Inspiron.
Post
When I launch an application it takes a pretty long time to be launched and I see a lot of disk read/write. read/wr ite. If the system was short on memory, this would be understandable as the system would use swap. But that's not the case in my situation (i.e. I have this problem even when the RAM is almost empty).
Question Candidates
Contains more than one good question!
1. How much much ram ram do you you have have installed installed ? and and what what size it the the swap swap disk partition partition ? 2.
If you do not have any problem with getting a little techy then may i suggest a method ?
3. How is it slow slow exactl exactly y ? boot time time ? hdd read/w read/write rite ? cpu time time ? graphics graphics rende rendering ring ? 4.
What is the longest time you have let it run ?
5.
This may be a silly question but ... did you make your usb stick bootable ?
6.
Do your system were recently updated ?
7.
Why not have two ssds in raid 1 for redundancy ?
8.
Is that a `parted -- list` on the synology device ?
9. Can you tell us us a little about about your your configura configuration tion ? 10. Did you you turn hardware virtualization on in bios/efi bios/efi ?
79
Evaluation
Evaluation set creation process o
We recruit 10 Unix admin experts using UpWork
o
Given a post and the set of ten question candidates "
Mark the one best question
"
Mark any other valid questions
80
Evaluation
Evaluation set creation process o
We recruit 10 Unix admin experts using UpWork
o
Given a post and the set of ten question candidates "
Mark the one best question
"
Mark any other valid questions
o
We annotate a total of 500 posts from our test set
o
Each post is annotated by two experts
81
Evaluation
Evaluation set creation process o
We recruit 10 Unix admin experts using UpWork
o
Given a post and the set of ten question candidates "
Mark the one best question
"
Mark any other valid questions
o
We annotate a total of 500 posts from our test set
o
Each post is annotated by two experts
o
Union of Bests: Questions Question s marked as best by by either of the annotators annotat ors
o
Intersection of Valids: Questions Questio ns marked marked as valid by both annotators
Best
Valid Valid
A1 A2 A1 A2
Union of Bests: { Q2, Q3 }
Q1 Q2
Intersection Intersec tion of Valids: { Q1, Q3, Q5 }
Q3 Q4 Q5 82
Evaluation
Baseline Models 1. Random: Randomly permute the 10 candidate questions
83
Evaluation
Baseline Models 1. Random: Randomly permute the 10 candidate questions 2. Ba Bagg-of of-n -ngr gram ams: s: Train linear classifier using bag-of-ngrams bag-of-ngrams of p, q and a
84
Evaluation
Baseline Models 1. Random: Randomly permute the 10 candidate questions 2. Ba Bagg-of of-n -ngr gram ams: s: Train linear classifier using bag-of-ngrams bag-of-ngrams of p, q and a 3. Co Comm mmun unit ity y QA (Nakov et al., 2017) : o o
o
SemEval Task: Rank comments by relevance relevance to post on Qatar Living Winning model: Logistic regression regression trained with string similarity & word embedding based features (Nandi et al., 2017) Our baseline: baseli ne:
We retrain this model on our dataset datase t
85
Evaluation
Baseline Models 1. Random: Randomly permute the 10 candidate questions 2. Ba Bagg-of of-n -ngr gram ams: s: Train linear classifier using bag-of-ngrams bag-of-ngrams of p, q and a 3. Co Comm mmun unit ity y QA (Nakov et al., 2017) : o o
o
SemEval Task: Rank comments by relevance relevance to post on Qatar Living Winning model: Logistic regression regression trained with string similarity & word embedding based features (Nandi et al., 2017) Our baseline: baseli ne:
We retrain this model on our dataset datase t
4. Neur Neural al (p (p,, q) Feedforward
p
qi
Post LSTM
Ques LSTM
Value V alue between between 0 and 1
Word embedding module p
qi 86
Evaluation
Baseline Models 1. Random: Randomly permute the 10 candidate questions 2. Ba Bagg-of of-n -ngr gram ams: s: Train linear classifier using bag-of-ngrams bag-of-ngrams of p, q and a 3. Co Comm mmun unit ity y QA (Nakov et al., 2017) : o o
o
SemEval Task: Rank comments by relevance relevance to post on Qatar Living Winning model: Logistic regression regression trained with string similarity & word embedding based features (Nandi et al., 2017) Our baseline: baseli ne:
We retrain this model on our dataset datase t
4. Neur Neural al (p (p,, q)
Value V alue between between 0 and 1
Feedforward
5. Neur Neural al (p (p,, q, q, a) a) pi
qi
Post LSTM
Ques LSTM
ai Ans LSTM
Word embedding module
pi
qi
ai 87
Evaluation
Baseline Models 1. Random: Randomly permute the 10 candidate questions 2. Ba Bagg-of of-n -ngr gram ams: s: Train linear classifier using bag-of-ngrams bag-of-ngrams of p, q and a 3. Co Comm mmun unit ity y QA (Nakov et al., 2017) : o o
o
SemEval Task: Rank comments by relevance relevance to post on Qatar Living Winning model: Logistic regression regression trained with string similarity & word embedding based features (Nandi et al., 2017) Our baseline: baseli ne:
We retrain this model on our dataset datase t
4. Neur Neural al (p (p,, q)
Value V alue between between 0 and 1
Feedforward
5. Neur Neural al (p (p,, q, q, a) a) pi
qi
Post LSTM
Ques LSTM
ai Ans LSTM
Both Neural (p, q, a) and EVPI (q | p, a) have similar no. of parameters
Word embedding module
pi
qi
ai 88
Evaluation
RESULTS Union of Best EVPI Neural (p, q, a) Neural (p, q) Community QA Bag-of-ngrams Random 0
Precision @1
89
Evaluation
RESULTS Union of Best EVPI Neural (p, q, a) Neural (p, q) Community QA Bag-of-ngrams
19.4
Random
17.5 0
5
10
15
20
25
30
35
40
Precision @1
90
Evaluation
RESULTS Union of Best EVPI Neural (p, q, a)
25.2
Neural (p, q) Non-linear vs linear Community QA Bag-of-ngrams
19.4
Random
17.5 0
5
10
15
20
25
30
35
40
Precision @1
91
Evaluation
RESULTS Union of Best EVPI Neural (p, q, a)
25.2
Neural (p, q)
21.9
Explicitly modeling “answer” is useful
Community QA Bag-of-ngrams
19.4
Random
17.5 0
5
10
15
20
25
30
35
40
Precision @1
92
Evaluation
RESULTS Union of Best EVPI Neural (p, q, a)
25.2
Neural (p, q)
21.9 Both use only (p, q)
Community QA
23.1
Bag-of-ngrams
19.4
Random
17.5 0
5
10
15
20
25
30
35
40
Precision @1
93
Evaluation
RESULTS Union of Best EVPI
27.7
Neural (p, q, a)
25.2
Neural (p, q)
21.9
Community QA
23.1
Bag-of-ngrams
19.4
Random
17.5 0
5
10
15
20
25
30
35
40
Precision @1 Diffe rence between EVPI and all baselines is statistically significant with p < 0.05 Note: Difference 94
Evaluation
RESULTS Union of Best EVPI
27.7
Neural (p, q, a)
Mainly differ in their loss function
25.2
Neural (p, q)
21.9
Community QA
23.1
Bag-of-ngrams
19.4
Random
17.5 0
5
10
15
20
25
30
35
40
Precision @1 Diffe rence between EVPI and all baselines is statistically significant with p < 0.05 Note: Difference 95
Evaluation
RESULTS Intersection Intersec tion of Valids
Union of Best
EVPI
36.1
27.7
Neural (p, q, a)
34.4
25.2
Neural (p, q)
31.6
21.9
Community QA
33.6
23.1
Bag-of-ngrams
25.6
19.4
Random
26.4
17.5
0
5
10
15
20
25
30
35
40
Precision @1 Diffe rence between EVPI and all baselines is statistically significant with p < 0.05 Note: Difference 96
Evaluation
RESULTS Intersection Intersect ion of Valids
Union of Best 32.2
EVPI
23.4 31.8
Neural (p, q, a)
22.7
Not statistically 30
Neural (p, q)
20.9
significant
30.8
Community QA
21.2 27.6
Bag-of-ngrams
19.4 26.4
Random
17.5
0
5
10
15
20
25
30
35
Precision @3
97
Talk Outline o
Clarification Questions Dataset
o
Problem Formulation: ormulatio n: Question Ranking
o
Expected Value of Perfect Information (EVPI) inspired model
o
Evaluation
o
Conclusion
98
CONCLUSION #
Key Contributions "
Create a dataset of ~77K clarification questions (and answers) with context
"
Introduce novel model that integrates deep learning with classic notion of expected value of perfect information
"
Create an evaluation set of size 500 with expert human annotations
99
CONCLUSION #
Key Contributions "
Create a dataset of ~77K clarification questions (and answers) with context
"
Introduce novel model that integrates deep learning with classic notion of expected value of perfect information
" #
Create an evaluation set of size 500 with expert human annotations
Key findings have multiple good clarification question " A context can have "
Explicitly modeling the answer helps in identifying good questions
"
EVPI formalism provides leverage over similarly expressive feedforward network
100
CONCLUSION #
Key Contributions "
Create a dataset of ~77K clarification questions (and answers) with context
"
Introduce novel model that integrates deep learning with classic notion of expected value of perfect information
" #
Create an evaluation set of size 500 with expert human annotations
Key findings have multiple good clarification question " A context can have
#
"
Explicitly modeling the answer helps in identifying good questions
"
EVPI formalism provides leverage over similarly expressive feedforward network
Future work "
Sequence-to-sequence based question generation generation model
"
Multi-turn question generation
"
How How to automatically evaluate performance?
101
CONCLUSION #
Key Contributions "
Create a dataset of ~77K clarification questions (and answers) with context
"
Introduce novel model that integrates deep learning with classic notion of expected value of perfect information
" #
Create an evaluation set of size 500 with expert human annotations
Key findings have multiple good clarification question " A context can have
#
"
Explicitly modeling the answer helps in identifying good questions
"
EVPI formalism provides leverage over similarly expressive feedforward network
Future work "
Sequence-to-sequence based question generation generation model
"
Multi-turn question generation
"
How How to automatically evaluate performance?
CODE + DATA: https://github.com/raosudha89/ranking_clarification_questions
102
Backup Slides
103
Evaluation
RESULTS Precision @ 1 Intersection Intersec tion of Valids
Union of Best
EVPI
36.1
27.7
Neural (p, q, a)
34.4
25.2
Neural (p, a)
32.3
24.1
Neural (p, q)
31.6
21.9
Community QA
33.6
23.1
Bag-of-ngrams
25.6
19.4
Random
26.4
17.5
0
5
10
15
20
25
30
35
40
104
Evaluation
RESULTS Precision @3 Intersection Intersec tion of Valids
Union of Best
EVPI
32.2
23.4
Neural (p, q, a)
31.8
22.7
Neural (p, a)
31.5
23.5
Neural (p, q)
20.9
Community QA
21.2
Bag-of-ngrams
26.4
17.5
0
5
10
15
30.8 27.6
19.4
Random
30
20
25
30
35
105
Evaluation
RESULTS Mean Average Precision Intersection Intersec tion of Valids
Union of Best
EVPI
49.2
43.6
Neural (p, q, a)
42.5
Neural (p, a)
46.5
41.4
Neural (p, q)
45.5
39.2
Community QA
40.2
Bag-of-ngrams
34.4
Random
35.2
0
10
20
30
47.7
47
42.7 42.1
40
50
60
106
Evaluation
RESULTS True Question EVPI
21.4
Neural (p, q, a)
20.5
Neural (p, q)
15.4
Community QA
18.5
Bag-of-ngrams
10.7
Random
10 0
5
10
15
20
25
30
35
40
Precision Diffe rence between EVPI and all baselines is statistically significant with p < 0.05 Note: Difference 107
Evaluation
RESULTS Union of Best (with true removed) EVPI
23.7
Neural (p, q, a)
22.2
Not statistically
Neural (p, a)
22.6
Neural (p, q)
significant
20.6
Community QA
22.6
Bag-of-ngrams
16.3
Random
17.4
0
5
10
15
20
25
30
Precision @1
108
SAMPLE OUTPUT Too much disk read/write when launching an application I have Xubuntu 13.04 on an old Dell Inspiron.
Post
When I launch an application it takes a pretty long time to be launched and I see a lot of disk read/write. read/wr ite. If the system was short on memory, this would be understandable as the system would use swap. But that's not the case in my situation (i.e. I have this problem even when the RAM is almost empty).
EVPI value
Ranking of Question Candidates
Best
Valid
0.21
How much ram do you have installed? and what size is the swap disk partition
0.18
Can you tell us a little about your configuration ?
0.17
What is the longest time you have have let it run ?
0.11
How is it slow exactly ? boot time ? hdd read/write ? cpu time ?
0.00
If you do not have any problem with getting a little techy may i suggest a method ?
0.00
This may be a silly question but ... did you make your your usb stick bootable ?
0.00
Do your your system were were recently recently updated ?
0.00
Why not have two ssds in raid 1 for redundancy redundancy ?
0.00
Is that a `parted -- list` on the synology synology device ?
0.00
Did you turn hardware hardware virtualization on in bios/efi ?
109
SAMPLE OUTPUT No wifi after restart in Ubuntu 16.04 After upgrading upgrading to 16.04, 16.04, there there is no wifi whene whenever ver I restart restart the system. system. My wireless wireless interface interfa ce of Ubuntu is RT3290 Wireless 802.11n 1T/1R PCIe On iwconfig I got the following eth0 no wireless extensions …. Currently to start wifi again I have to shutdown, then boot the system again. How to fix the problem?
EVPI value
Ranking of Question Candidates
Best
Valid Valid
0.24
I doubt it, shutdown and reboot are exactly identical! are you really rebooting?
0.13
Be clear about the problem. Is Ubuntu not showing them even though they are present?
0.11
What is 4g wifi connection? connection?
0.09
Can you type `iwconfig` in terminal and paste what it returns here?
0.09
What does this tell us?
0.08
If I post it as an answer, would you kindle mark as such?
0.06
Which Which Ubuntu Ubuntu 15?
0.06
What exactly exactly do you you mean by make fails?
0.05
Welcome to ask Ubuntu! ; - ) Is the wireless lan disabled in the bios?
0.00
Is Ubuntu detecting your wireless card ? **iwconfig** does list your card?
110
Expected Value of Perfect Information (EVPI) inspired model
LONG SHORT TERM MEMORY (LSTM)
Sepp Hochreiter and J¨urgen Schmidhuber. 1997. Long short-term memory. Neural computation , 9(8):1735–1780. Jeffrey Jeffrey Pennington, Richard Socher, Socher, and Christopher D Manning. 2014. “Glove: Global vectors for word word representation” representation” In Empirical Methods on Natural Language Processing. 111