Popularity in Informed Decentralized Search Denis Helic & - - PowerPoint PPT Presentation

β–Ά
popularity in informed
SMART_READER_LITE
LIVE PREVIEW

Popularity in Informed Decentralized Search Denis Helic & - - PowerPoint PPT Presentation

Aug 26, 2022 273 likes β€’502 views

1 The Role of Homophily and Popularity in Informed Decentralized Search Denis Helic & Florian Geigl Knowledge Technologies Institute Infeldgasse 13/5. floor, 8010 Graz, Austria {florian.geigl,dhelic}@tugraz.at http://kti.tugraz.at/

slide-1
SLIDE 1

u www.tugraz.at

The Role of Homophily and Popularity in Informed Decentralized Search

September 15, 2014 Denis Helic & Florian Geigl

1

Knowledge Technologies Institute Infeldgasse 13/5. floor, 8010 Graz, Austria {florian.geigl,dhelic}@tugraz.at http://kti.tugraz.at/

slide-2
SLIDE 2

The Role of Homophily and Popularity in Informed Decentralized Search

  • Decentralized Search
  • Informed Decentralized Search
  • steered by some kind of knowledge

September 15, 2014 Denis Helic & Florian Geigl 2

Start Target

slide-3
SLIDE 3

The Role of Homophily and Popularity in Informed Decentralized Search

September 15, 2014 Denis Helic & Florian Geigl 3

Start Target

Popularity Homophily

slide-4
SLIDE 4

Motivation

  • large networks
  • dynamic networks
  • no central search
  • P2P
  • swarm of drones

September 15, 2014 Denis Helic & Florian Geigl 4

Stackoverflow.com Communication-Network

slide-5
SLIDE 5

Related Work

September 15, 2014 Denis Helic & Florian Geigl 5

Start Target

Adamic Kleinberg fixed mixture: Jensen

slide-6
SLIDE 6

Proxies

Homophily: cosine similarity to target node

0 <= cosine similarity <= 1

π‘‘π‘π‘›π‘›π‘π‘œ π‘œπ‘“π‘—π‘•β„Žπ‘π‘π‘£π‘ π‘‘(𝑗,π‘˜) 𝑒𝑓𝑕𝑠𝑓𝑓 𝑗 βˆ—π‘’π‘“π‘•π‘ π‘“π‘“(π‘˜)

Popularity: degree of the node

September 15, 2014 Denis Helic & Florian Geigl 6

slide-7
SLIDE 7

Example: Greedy Navigation using Popularity

September 15, 2014 Denis Helic & Florian Geigl 7

Start Target 4 8

slide-8
SLIDE 8

Normalization

September 15, 2014 Denis Helic & Florian Geigl 8

Start Target 4 12 8 12

slide-9
SLIDE 9

Mixture Distribution

September 15, 2014 Denis Helic & Florian Geigl 9

mixture = p*Ξ± + q*(1- Ξ±)

slide-10
SLIDE 10

Mixture Distribution

September 15, 2014 Denis Helic & Florian Geigl 10

H P

slide-11
SLIDE 11

Datasets

  • DBLP
  • Facebook Subset
  • Twitter Subset
  • Wikipedia for

Schools nodes ~4k – ~300k

September 15, 2014 Denis Helic & Florian Geigl 11

slide-12
SLIDE 12

Experimental Setup & Evaluation

random missions vary Ξ± from 0 to 1 Success Rate

September 15, 2014 Denis Helic & Florian Geigl 12

slide-13
SLIDE 13

Results Greedy Navigation

September 15, 2014 Denis Helic & Florian Geigl 13

mixture: H*Ξ± + P*(1- Ξ±)

H P

slide-14
SLIDE 14

Background Knowledge Models

  • static mixture βœ”
  • static switch
  • inspired by human navigation

September 15, 2014 Denis Helic & Florian Geigl 14

Start Step 1 Step x Step n-1 Target

Ξ± = Initial Ξ± Ξ± = 1 - (Initial Ξ±)

slide-15
SLIDE 15

Results Greedy Navigation

September 15, 2014 Denis Helic & Florian Geigl 15

slide-16
SLIDE 16

Background Knowledge Models

  • dynamic switch

September 15, 2014 Denis Helic & Florian Geigl 16

Start Step 1 CosSim not uniform Step n-1 Target

Ξ± = Initial Ξ± Ξ± = 1 - (Initial Ξ±)

slide-17
SLIDE 17

Results Greedy Navigation

September 15, 2014 Denis Helic & Florian Geigl 17

mixture: H*Ξ± + P*(1- Ξ±)

H P

slide-18
SLIDE 18

Navigation Models

  • greedy search
  • always use best
  • stochastic search
  • draw out of mixture distribution
  • softmax search:
  • apply softmax on convex combination
  • draw out of resulting distribution

September 15, 2014 Denis Helic & Florian Geigl 18

slide-19
SLIDE 19

Softmax

September 15, 2014 Denis Helic & Florian Geigl 19

slide-20
SLIDE 20

Results Stochastic & Softmax

September 15, 2014 Denis Helic & Florian Geigl 20

slide-21
SLIDE 21

Discussion

  • Homophily seems to be more important
  • degree distribution
  • low diameter networks
  • cosine similarity includes

a lot of information

September 15, 2014 Denis Helic & Florian Geigl 21

effective diameter

slide-22
SLIDE 22

When searching your β€žnodeβ€œ, donβ€˜t pick the popular ones, take the similar 

September 15, 2014 Denis Helic & Florian Geigl 22