Marc Smith Microsoft Live Community Research [email protected].
On Publication of the Research - microsoft.com...Outline Why publishing a research paper Category of...
Transcript of On Publication of the Research - microsoft.com...Outline Why publishing a research paper Category of...
On Publication of the Research
Dr. Yu Zheng
Researcher @ Web Search & Mining Group
Microsoft Research Asia
http://research.microsoft.com/en-us/people/yuzheng/
A reminder
• Do not treat my experiences as rules
• Do better using your intelligence and creativity
Outline
Why publishing a research paper
Category of research publications
Styles of different conferences
How to evaluate a research paper
Art of the writing
Warnings
……. …... …… ……
……. …... …… ……
……. …... …… ……
……. …... …… ……
……. …... …… ……
…….. …… ……
Why publishing a research paper
Contribute to the community
Propose new problem and research directions
Share your idea and methodologies to others
Feedback and improve other people’s work
Educate and knowledge new-comers to the community
Benefit from publications
Well document your work, shape your thoughts…
Practice presentation skill
Get feedback from others
Connect to peers and identify collaboration
Build personal credit in the community
Respect, honor, and awards……
Contribution
Benefit
Healthy Career
Category of research publications
Technical report
Workshop publications
Conference publications Full/oral papers
Short paper
Posters
Demos
Videos
Doctoral Colloquium
On the progress work
Tutorial
Journal publications Full research articles
Regular
Special issues
Review and comment papers
Book
Preliminary idea
-Workshop
Shaped work
-Conference
Feedback
Effo
rt
Period
Feedback
Expanded work
-Journal Feedback
Summarized work
-Book Feedback
Styles of different conferences
• Human and Interfaces – CHI, UIST, …
• Applications and systems – WWW, SIGIR, …
– Ubicomp, Pervasive, Percom…
– KDD industrial track
– Mobisys, ACM MM…
• Theory and models – SIGMOD, VLDB, ICDE, KDD research
track…
– AAAI, IJCAI…
– ICML, NIPS…
Human and interaction Applications and systems
Theory and models
Language style Natural Natural + Formal Formal
Scales of Experiment Rich Middle Small
Experiment styles User study/in the field
study Synthetic test/lab. test
/user study Synthetic test/lab. test
Lemma and proof Rare Middle Rich
How to evaluate a conference paper
Relevance to a conference or journal
Novelty and originality
Significant contribution Insight
Strong points
Technical sound Methodology
experiments
Presentation Clear structure
Good readability
Proper length and details
Related work study
The structure of a research paper
Introduction
Related work
Methodology
Experiments
Discussion
Conclusion & Future work
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
• Thoughtfully beautiful pictures
• novel
• Insight
• Deliver key messages
• Hierarchical structure
• Skip some details
• revise the paper
• Attractive appearance
• Original
• Soul
• Highlight Key features
• Step-by-step carving
• Ignore some pores
• Polish a sculpture
……. …... …… ……
Introduction - The most difficult part to write
Background
Goal
Motivation Highlight the key insight
Benefit and application scenarios
Difficulties and challenges
Brief your method
Contributions
Structure of the rest
Background
Motivation
Goal
Insight Challenges
Introduction
Contribution How well it works
Support and validation
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
Hierarchical Structure
a b c d e f g h
a b c d e f g h
A C B
R
Road Networks
Time-Dependent
Landmark Graph
Trace Preprocessing Landmark Graph Construction Route Computing
Trace Archive
GPS Logs
Trace Segmentation Landmark Graph Building
User Query
Rough Routing
Refined Routing
User Interface
Travel Time Estimation
Map Matching
A Time-dependent
Landmark Graph
Taxi Trajectories
A Road Network
Rough
Routing
Refined
Routing
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
Related work
Categorize literatures
Differentiate your work
Emphasize contributions
Justify the novelty
Position your work
Position in a paper
Right after introduction
Right before conclusion
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
Clear definition
Define it until you need it
Use figures to demonstrate
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
Thoughtfully beautiful figures
Substantiate vision and concept
Presenting framework and architecture
Illustrating methodology
Showing results
Difficult
Easy
Service providing
1. Send a query
Q=(qs, qd, t, α)
Weekend
Severe weather
Weekday
3. Route downloading
4. Logging the
real travel with
a GPS trace
5. Learning
new α
2. Route
computing Offline
mining
Online
inference
Real-time
taxi trajectories
Ph
ysi
cal
wo
rld
Cy
ber
wo
rld
Knowledge discovery
Normal weather
Landmark Graphs
Real/Future traffic
Historical trajectories
and weather
Landmark Graphs
A Time-dependent
Landmark Graph
Taxi Trajectories
A Road Network
Rough
Routing
Refined
Routing
Substantiate idea and concept
GeoLife: A Location-History-Based Social Network
26
POI/YP DB
Presenting framework
Modularization
Mapping to paper sections
Road Networks
Time-Dependent
Landmark Graph
Trace Preprocessing Landmark Graph Construction Route Computing
Trace Archive
GPS Logs
Trace Segmentation Landmark Graph Building
User Query
Rough Routing
Refined Routing
User Interface
Travel Time Estimation
Map Matching
Presenting framework
Modeling Location History
HITS-Based Inference Model
Mining Classical Travel Sequences
GPS Logs
GPS-Phones
Location
Interest
User Travel
Experience
Classical Travel
Sequences
Interesting Locations
Experienced Users
Tree-Based
Hierarchical Graph
Location Recommender
Laptops and PCs
Mobile
Communicator
Location Interest and Sequence MiningKnowledge
Recommendation Location History Modeling
Road Networks
Time-Dependent
Landmark Graph
Trace Preprocessing Landmark Graph Construction Route Computing
Trace Archive
GPS Logs
Trace Segmentation Landmark Graph Building
User Query
Rough Routing
Refined Routing
User Interface
Travel Time Estimation
Map Matching
Change Point
Clustering
Training Data
Segmentation
Extracting
Feature
Knowledge
Extraction
Test Data
Segmentation
Extracting
Feature
Post-Processing
Spatial
Knowledge
Model Training Inference
Model
Trans.
ModesSpatial Indexing
Spatially Indexed
Knowledge
Online Inference Offline Learning
Graph Building
Trip Partition
Trajectory Archive
Map Matching
Indexing
Low-sampling-rate Query Trajectory
Simple Reference
Search
Spliced Reference
Search
Reference Trajectories
Graph based
Approach
Nearest Neighbor
based Approach
Route Scoring
Global Route Inference
Reference Search
Local Route Inference
GPS Logs
Preprocessing Route Inference
R-tree Index
Candidate Edges
Illustrating methodology
2γ
Geographic
spaces
Feature
spaces
Feature space
Hierarchical
clustering
Location category hierarchy
C1
C2
C3
Layer 2
Clusters at a layer
Location history of User 1 Location history of User m
Feature space
Layer 3
Layer 1
Feature space
Layer 2
Layer 3
Layer 1
User 1
Feature vectors of stay regions Feature vectors of stay regions
User m
2γ
a stay point
a stay region
a feature vector
c32
c31
c30
c20 c21
Tram
1 0 0 0 0 0
0 1 0 0 0 0
0 0 0 1 0 0
0 0 1 0 0 0
0 0 0 0 1 0
0 0 0 0 0 1
A B C D E F
A
B
D
C
E
F
6A11
5
B224C34 3 D43
2E55
1 F66
A
redundant
(a) The match matrix (b) The precedent graph
0 0 0 0 0 0G7
1
2
3
4
5
6
61 2 3 4 5
A trivial
match
A precedent
relationship
Index
7 9 14 17 19 24
0.2
0.4
0.6
0.8
1
time of day (hour)
pro
po
rtio
n
3-5min
5-10min
10-14min
r2
Tr1 r3
r9
r8
r6
r1
Tr2
Tr5
Tr3
Tr4
A) Matched taxi trajectories B) Detected landmarks C) A landmark graph
r9
r3r1
r6
r9
r3r1
r6
p1 p2
D) Travel time estimation
p3 p4
r4
r5r7
r10
e16
e96
e93
e13
e637 11 16 19 21 24
0.2
0.4
0.6
0.8
1
time of day (hour)
pro
po
rtio
n
2-4min
4-9min
e13
e16
Figures showing results
Curve
Bar
Pie
Table
Visualization 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Decision
Tree
SVM Bayes net CRF
Infe
ren
ce A
ccu
racy
Inference Model
Accuracy by Length
Accuracy by Duration
100 200 300 400 500
0.646
0.684
0.722
0.760
Acc
ura
cy
Number of training taxis
DT+HSMM
DT
HMM
CRF
16%
45%
30%
9%
age<=22 22<age<=25
26<=age<29 age>=30
Figures showing results
Visualization
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
Thorough and persuasive evaluation
Settings Datasets
Parameters
Users and objects
Approach Strategy
Metrics
Ground truth
Baselines
Results Data conclusion
Analysis and reasons
Compare with baselines
Self-exploration
Writing is an art
Charming introduction Interesting and clear goal
Substantial motivation and challenges
Significant contribution
Hierarchical structure
Well positioned related work
Clear definitions
Thoughtfully beautiful figures
Thorough and persuasive experiments
Clearly claimed conclusion
Clearly claimed conclusion
Emphasize contribution and novelty
Highlight key results with numbers
Be careful using auxiliary verb Weak: can, may, might, would, could…
Strong: did, do, will
How to present a component
Step-wise
WWHW What to do
Why
How to
Why
Using figures and running examples
What: detect parking places
Why using the method: traffic jams and K-mean does not work
Why: segment trajectories and reduce complexity
How to
Warnings
Many details: A research paper is not a specification
Missing key messages
Presenting a simple problem in a complex way
Use too many symbols
Criticize other people’s work
Over claim and excessive contribution
Take away
People
Structure
Figures
Related work
……. …... …… ……
……. …... …… ……
……. …... …… ……
……. …... …… ……
……. …... …… ……
…….. …… ……
Thanks!
Yu Zheng, Web Search & Mining Group
http://research.microsoft.com/en-us/people/yuzheng/