complexity resolution control for context based on metromaps

17
Marat Zhanikeev [email protected] maratishe.github.io 2016/01/22@LOISFIT Complexity Resolution Control PDF: bit.do/160122 for Context Based on Metromaps #metromaps #visualization #complexity #searchspace

Upload: kyushu-institute-of-technology

Post on 18-Feb-2017

383 views

Category:

Technology


1 download

TRANSCRIPT

Marat [email protected]/01/22@LOIS研@FIT

Complexity Resolution Control

PDF: bit.do/160122

for Context Based on Metromaps

#metromaps#visualization #complexity #searchspace

.

The Metromap

07 K.Nesbitt "Getting to more abstract places using the metro map metaphor" 8th IV (2004)

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 2/17...

2/17

.

Metromaps in Practice

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 3/17...

3/17

.

In Practice: Context Management + AI

Rebot

(careless) Input

Human Human

{structure}

(pinpoint) Select

Browse (or use otherwise)

Some Knowledge

(folksonomies, knowledge bases, databases, indexes, ontologies, etc.)

(metromaps )

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 4/17...

4/17

.

Metromaps: Lines and Stations

01 M.Zhanikeev "Multidimentional Classification Automation with Human Interface based on Metromaps" 4th AII (2015)M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 5/17

...

5/17

.

Data for this Study (tourism)• comes from real Big Data -- not supposed to share the source, yet public

• tourism stats: location, category (culture, food, ...), class (local/inbound), age• blue line connects randomly selected stations on each data line

location

category age

class

location

category

age

class

add

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 6/17...

6/17

.

Complexity Resolution

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 7/17...

7/17

.

Concepts and Connections

• applied to decisions which have to deal with a large search space◦ big data analysis, robotics, social systems

• traditional ontologies→metromaps• complexity resolution control = growing/shrinking area around the mainstation on the metromap

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 8/17...

8/17

.

Metromap and Distribution Curves

0 5 10 15 20 25 30 35 40 45Decreasing order

4.55

4.9

5.25

5.6

5.95

6.3

6.65

7

7.35

7.7

8.05

log(

1 +

v)

• the same tourism bigdata

• each curve is distributionof mass acrossvalues=stations

• distributions are often withhotspots 02

• metromap is a networkthat overlays the curves

02 M.Zhanikeev "The Next Generation of Networks is all about Hotspot Distributions and Cut-Through Circuits" IEICE・CQ研 (2015)

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 9/17...

9/17

.

SameLine and AnyLine Models

B

Line 1 Line 2 Line 3

A

Same -line, 1 station away (from A)

Any line, 2 stations away (from B)

• same line : can includeother stations on the same line,but not transfer to another line

• any line: can transfer toother lines

• note: the assumption is thatsome stations are shared bytwo or more lines

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 10/17...

10/17

.

Results

0 10 20 30 40 50

Increasing sequence

5.6

5.95

6.3

6.65

7

7.35

7.7

log(

1 +

gro

wth

)

Cross#2 Hops#1

0 20 40 60

Increasing sequence

6.3

6.65

7

7.35

7.7

8.05

8.4

log(

1 +

gro

wth

)

Cross#2 Hops#2

0 10 20 30 40 50

Increasing sequence

7

7.5

8

8.5

log(

1 +

gro

wth

)

Cross#2 Hops#3

SamelineAnyline

0 20 40 60

Increasing sequence

5.6

6

6.4

6.8

7.2

7.6

log(

1 +

gro

wth

)

Cross#3 Hops#1

0 20 40 60

Increasing sequence

6.3

6.65

7

7.35

7.7

8.05

8.4

log(

1 +

gro

wth

)

Cross#3 Hops#2

0 20 40 60

Increasing sequence

6.3

6.65

7

7.35

7.7

8.05

8.4

log(

1 +

gro

wth

)

Cross#3 Hops#3

• aggregated data:number oftourists

• each column is fora given number ofhops from a randomlyselected station

• each row is for thenumber of intersectinglines

• beware the log scale

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 11/17...

11/17

.

On the Other Side...

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 12/17...

12/17

.

Inverse Problem: Growing Search Space

0 5 10 15 20 25 30 35 40 45Decreasing order

4.55

4.9

5.25

5.6

5.95

6.3

6.65

7

7.35

7.7

8.05

log(

1 +

v)

• the same input: metromapoverlaying distributioncurves

• different objective: maximizecount increment for aminimum increase instation coverage

• translated into tourism: how tomaximize increase innumber of tourists atminimal cost

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 13/17...

13/17

.

Inverse Problem Results• response to unit cost, adding only one station on a randomly selected line

0 2.65 5.3 7.95 10.6 13.25 15.9 18.55 21.2 23.85Decreasing order

0200400600800

10001200140016001800

Ad

ded

val

ue (x

10,0

00 p

eopl

e)agecategorylocationclass

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 14/17...

14/17

.

That’s all, thank you ...

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 15/17...

15/17

.

More on Data and Metromaps

Kind of tourism

Age

Inbound(inside)

Japan

未成年 若年層 中年層 老年層

Place北海道 東京都 京都府 福岡県 ……

Kind of activity

食べる 温泉 ……

Young people who come from abroad to Tokyo on a food tour.

Tourist counts onsimilar tours toSapporo is much smaller

What is the cost/benefitof connecting these?

• if you plot only the raw = basebigdata, then lines never intersect --such a metromap is no fun

• the point is to connect all lines byselecting a specific case = kind oftourist

• note: this is one of many ways togenerate metromaps in practice

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 16/17...

16/17

.

You Know This as "Social Media"• I call it head biting tail model, means: popular items connect to rarelyvisited ones

Decreasing order

You can try to lift the tail by itself... But it is

much easier to connect head with tail.

M.Zhanikeev -- [email protected] Complexity Resolution Control for Context Based on Metromaps -- bit.do/160122 17/17...

17/17