on context management using metro maps

24

Upload: marat-zhanikeev

Post on 12-Jul-2015

196 views

Category:

Technology


0 download

TRANSCRIPT

Page 1: On Context Management Using Metro Maps
Page 2: On Context Management Using Metro Maps

.

Metromaps: Why the Fuss?

• we want to manage our context but in folksonomy style (people are

lazy and sloppy)

• interactive browser visualizations would be nice

• we want to personalize our own context (personal, teams, companies,schools, etc.)

• we want robots to create and manage our context

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 2/24...

2/24

Page 3: On Context Management Using Metro Maps

.

Facets: the Main Rival

.Facets are.....

.... a means of putting tags in context, thus reducing ambiguity

• synonyms: aspects, domains, layered context, ...

• major merit: one can browse context by facets

• major flaw: too crude• visual merit: messes up visualizations by destroying the original structure

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 3/24...

3/24

Page 4: On Context Management Using Metro Maps

.

Metromaps : the Basic Concept• the ordinary metro/subway/... maps, trains represent end-to-endpaths in context

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 4/24...

4/24

Page 5: On Context Management Using Metro Maps

.

Metromaps: Basic Semantics• same information, different visualization approaches, which one do you prefer?

(rhetorical)

marat

tim

takashi

status: marriedage: 36

status: single

age: 41

age: 38

married single

36 38 41

marat takashi tim

status line

age line

friendsline

maratline takashi

linetimline

MetromapOntology

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 5/24...

5/24

Page 6: On Context Management Using Metro Maps

.

Search : Metromaps vs Facets

Hits Search string

Search string

Facets

Hits Trains

Search

Ride

Select

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 6/24...

6/24

Page 7: On Context Management Using Metro Maps

.

Context : Input

Content

text2tags

tags2ontology

search

text2tags

doc2lines

metromap

lines stations

MetromapsOntology

Annotator

User

Annotate

Process

Browse

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 7/24...

7/24

Page 8: On Context Management Using Metro Maps

.

Context : Output

Fuzzytags

Walkontology

tree

Line(s)

Tag(s)

Station(s)

Connectby line(s)

Browsedoc list

AND/OR

Deeper until hit

Visualize(draw map)

Visual browsing

Inspect the list

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 8/24...

8/24

Page 9: On Context Management Using Metro Maps

.

Models: Walk Tree (ontology).Walk Tree Model..

.

Search all docs for tag X and all tags Y-hopsaway from X.

Fuzzytags

Walkontology

tree

Line(s)

Tag(s)

Station(s)

Connectby line(s)

Browsedoc list

AND/OR

Deeper until hit

Visualize(draw map)

Visual browsing

Inspect the list

• browse from tag to tagfollowing relational links

• the search area can be definedas number of hops awayfrom the first tag

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 9/24...

9/24

Page 10: On Context Management Using Metro Maps

.

Models: Fuzzy Tags (ontology).Fuzzy Tags Model..

.

Select several tags (from a list) to search fordocs with (at least one of) these tags.

Fuzzytags

Walkontology

tree

Line(s)

Tag(s)

Station(s)

Connectby line(s)

Browsedoc list

AND/OR

Deeper until hit

Visualize(draw map)

Visual browsing

Inspect the list

• OR mergers are morecommon

• the AND is bad for fuzzytags, search list will quicklyproduce ZERO hits

• AND problem can be improvedwith tagging discipline, butwe are in a folksonomy

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 10/24...

10/24

Page 11: On Context Management Using Metro Maps

.

Models: Line Hopping (metromaps).Line Hopping..

.

In metromaps, update search lists by jumping fromone train to another

Fuzzytags

Walkontology

tree

Line(s)

Tag(s)

Station(s)

Connectby line(s)

Browsedoc list

AND/OR

Deeper until hit

Visualize(draw map)

Visual browsing

Inspect the list

• visualization : tag →line A → other (connected)lines

• stations are tags (linesaggregate stations)

• when browsing, you get the listof docs for all stations incurrent line

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 11/24...

11/24

Page 12: On Context Management Using Metro Maps

.

Models: Line Intersection (metromaps).Line Intersection...Search docs on intersections of several lines

Fuzzytags

Walkontology

tree

Line(s)

Tag(s)

Station(s)

Connectby line(s)

Browsedoc list

AND/OR

Deeper until hit

Visualize(draw map)

Visual browsing

Inspect the list

• visualization: select severallines and get the doc list

• AND works fine herebecause lines arelarge-scale aggregates

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 12/24...

12/24

Page 13: On Context Management Using Metro Maps

.

Real Trace : 10+ years of Metadataha

veus

edip

dum

pps

amp

mym

2msa

mpl

ing

traffi

cte

qos

onof

fpsa

mp

mul

tisou

rces

tream

ing

myc

dn

stre

amin

gm

igra

tiona

void

ance cd

nha

vere

adm

2md

past

atra

ces

wor

kloa

dm

igra

tion

hybr

idp2

pw

ebsit

ep2

pka

shik

oibi

ttorre

ntflo

ws

0

20

40

60

80

Occ

urre

nce

(tim

es)

icc

glob

ecom

have

used

info

com itc

sprin

ger

tele

com

mun

icat

ion

syst

ems

2007 ic

t20

09ip

dum

pps

amp

2010

2008

nom

stm

mym

2mne

twor

king

aggn

new

sjn

smte

qos

sam

plin

gtra

ffic

0

60

120

180

240

300

360

420

Occ

urre

nce

(tim

es)

• top: a smaller repository for a separate project

• bottom: the entire metadata

• 10+ years of metadata managed acrossseveral labs

• a folksonomy: many people people, carelessmanagement style, etc.

• tags are also careless, mostly practical likehaveused (in a paper)

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 13/24...

13/24

Page 14: On Context Management Using Metro Maps

.

Results: The Four Modes

0 2 4 6 8Browsing round

0

0.45

0.9

1.35

1.8

2.25lo

g( n

o. o

f pap

ers)

Ontology: Walk Tree Model

0 2 4 6 8Browsing round

0

0.45

0.9

1.35

1.8

2.25

log(

no.

of p

aper

s)

Metromap: Train Hopping

random curves average (width=number of hits)

0 5 10 15 20 25Tag Grouping

0

0.45

0.9

1.35

1.8

2.25

log(

no.

of p

aper

s)

Ontology: Fuzzy Tags Model

0 5 10 15 20 25Line Intersections

0

0.45

0.9

1.35

1.8

2.25

log(

no.

of p

aper

s)

Metromap: Train Intersection

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 14/24...

14/24

Page 15: On Context Management Using Metro Maps

.

Current Work

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 15/24...

15/24

Page 16: On Context Management Using Metro Maps

.

Current Work

• ... continues in several directions

1. structure: order of stations in lines, graph optimization2. visualization: specifically interactive metromaps as

webapps

3. automation: recommendation bot and context management

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 16/24...

16/24

Page 17: On Context Management Using Metro Maps

.

Viz: Graphviz libraryviz

multidimensional

dependency

models

pcapca

softwaresoftwaresoftwaresoftware

kbseken

visualization

test

smil

performance

multimedia

modeling

todotodotodotodotodo

benchmark

testing

session

vne

ieiceconf

theory

crosslayer

game

ospf

cloud

optimization

opportunetstic

multiflow

networking

aggregationaggregation

hardware

opportunistic

p2pwifi

drive

connectivity

multiple

access

wirelesswireless

rcskenrcsken

virtual

content

direct

google

wifi

adhoc

establishment

aodv

differential

efficiency

budget

cost

backup

depletion

energy

e2e

path

battery

tomo

delay

coordinates

networks

values

missing

matrix

network

endtoendendtoendendtoendendtoendendtoendendtoendendtoend

tomography

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 17/24...

17/24

Page 18: On Context Management Using Metro Maps

.

Viz : (own) Interactive Webapps

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 18/24...

18/24

Page 19: On Context Management Using Metro Maps

.

Viz : (own) Circular Layouts

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 19/24...

19/24

Page 20: On Context Management Using Metro Maps

.

Rebot: A REcommendation BOT• Bayesian classification works better when context is aggregated bymetromap trains

• higher classification quality helps with practical recommendations

Facets Lucene query

List Facets

ToS Lucene query

List Trains

Recommendations

Robot

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 20/24...

20/24

Page 21: On Context Management Using Metro Maps

.

That’s all, thank you ...

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 21/24...

21/24

Page 22: On Context Management Using Metro Maps

.

Where is My Stuff?

Trains of

Sought

Search

Faceted Search

scalable

browsable scalable few

browsable teamable ownable

evolvable M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 22/24

...

22/24

Page 23: On Context Management Using Metro Maps

.

Automation: Rebot versus Others• rebot: REcommendation BOT• multiple items: metromap trains can aggregate otherwise unrelated papers

(on demand)

Multiple Items? Visual?

Underlying Structure? Learning?

Traditional Search NO NO NO NO

Facetted Search NO NO YES NO

Advanced Tools / Others NO YES YES YES

Rebot (the proposal) YES YES YES YES

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 23/24...

23/24

Page 24: On Context Management Using Metro Maps

.

Rebot : Social Robotics in Software• social robotics: a hardware topic where robots learn tasks by interacting withhuman guides

• context rebot: roughly the same concept implemented in software and appliedto context management

Generic Use

Teaching, Guidance Reasoning

Human Role

Social Robotics

Wide range of behavior

Reinforced Learning

YES. Vision, recognition Guide only

The Rebot (proposal)

Any kind of context

Bayesian Classification

NO Not needed

Guide and decision - maker

M.Zhanikeev -- [email protected] -- On Context Management using Metro Map -- http://bit.do/marat141119 -- 24/24...

24/24