Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
ネットワーク科学 中心性とGoogleのPageRank
Search
hayashilab
July 03, 2020
Science
0
24
ネットワーク科学 中心性とGoogleのPageRank
hayashilab
July 03, 2020
Tweet
Share
More Decks by hayashilab
See All by hayashilab
ネットワーク科学 ネットワークが社会を支える
hayashilab
0
32
ネットワーク科学 小さな世界のモデル
hayashilab
0
33
ネットワーク科学 蔓延する利己主義とScale-Free則
hayashilab
0
35
ネットワーク科学 構造と伝わりやすさ
hayashilab
0
24
ネットワーク科学 社会インフラのレジリエンス
hayashilab
0
41
ネットワーク科学 空間システムデザイン
hayashilab
0
31
ネットワーク科学最前線2017 -インフルエンサーと機械学習からの接近-
hayashilab
1
18
Other Decks in Science
See All in Science
BigQueryで参加するレコメンドコンペ / bq-recommend-competition-kaggle-meetup-tokyo-2023
shimacos
1
1.3k
DEIM2024 チュートリアル ~AWSで生成AIのRAGを使ったチャットボットを作ってみよう~
yamahiro
3
630
Mastering Feature Engineering: Mining the Hidden Salary Formula with CakeResume
tlyu0419
0
130
マルチモーダルモデルと自動運転 車載モデルのコスト・スループット・レイテンシ / LLM in Production Meetup #2 20231023
yuyamaguchi
1
1k
研究・教育・産学連携の循環の実践
sshimizu2006
0
220
SCOTT: Self-Consistent Chain-of-Thought Distillation
meshidenn
0
310
Machine Learning for Materials (Lecture 7)
aronwalsh
0
730
AI科学の何が“哲学”の問題になるのか ~問いマッピングの試み~
rmaruy
1
1.2k
Transformer系機械学習モデルを取り巻くライブラリや用語を整理する
bobfromjapan
2
480
同じデータでもP値が変わる話/key_considerations_in_NHST
florets1
1
1.1k
データで課題を解決する -因果関係を調べる統計的因果推論-
sshimizu2006
4
1.3k
[NeurIPS 2023 論文読み会] Wasserstein Quantum Monte Carlo
stakaya
0
350
Featured
See All Featured
Large-scale JavaScript Application Architecture
addyosmani
504
110k
jQuery: Nuts, Bolts and Bling
dougneiner
59
7.1k
Thoughts on Productivity
jonyablonski
58
3.8k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
125
32k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
187
16k
Build The Right Thing And Hit Your Dates
maggiecrowley
24
2k
What's in a price? How to price your products and services
michaelherold
237
11k
The Power of CSS Pseudo Elements
geoffreycrofte
60
5k
Typedesign – Prime Four
hannesfritz
36
2.1k
YesSQL, Process and Tooling at Scale
rocio
164
13k
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
20
1.9k
Designing on Purpose - Digital PM Summit 2013
jponch
110
6.5k
Transcript
Google PageRank 2 2019 ( ) Google PageRank 1 /
23
1. , , ( ) Google PageRank 2 / 23
, , , 6 , , 2007 ( ) Google
PageRank 3 / 23
2. (Degree Centrality): i ki ki /(N − 1) .
N − 1 , i N − 1 . , . (Information Centrality): i j ( ) . ( ) Google PageRank 4 / 23
(Closeness Centrality): i j : +1 Hij , j Hij
N − 1 −1 = N − 1 j Hij . . N − 1 . , . (Flow Centrality): i . ( ) Google PageRank 5 / 23
(betweenness centrality): v s,t δst (v), σst (s, t) ,
σst (v) v δst (v) def = σst (v) σst L.C. Freeman, Sociometry 40, 1977 http://www.geocities.jp/woodone3831/kanntou/c-4-11-sekisyo-HAKONE.html ( ) Google PageRank 6 / 23
BC http://astamuse.com/ja/published/JP/No/2010141442 ( ) Google PageRank 7 / 23
Brandes BC s u σsu , δs,•(w) w ∼ w′
t , w ∼ w′ t δs,•(v) = {w|v∈Ps (w)} σsv σsw (1 + δs,•(w)), Ps (w) def = {v ∈ V |(v, w) ∈ E, d(s, w) = d(s, v) + 1}, w ∈ ∂v s v w w’ Ps(w) t t’ : : : : σ sw σ sv : U.Brandes, Journal of Math. Sociology 25, 2001 ( ) Google PageRank 8 / 23
, δst (v) = u∈Preds , t (v) δst (u)
× R(s, u, v, t), Preds,t (v) s-t v 1 {u}, R(s, u, v, t) s t (u, v) , . s-t T(s, t) , v ∈ V . δ•,•(v) = s,t∈V δst (v) × T(s, t). S.Dolev et al., Journal of the ACM 57, 2010 ( ) Google PageRank 9 / 23
R(s , u , v , t) T(s , t)
s u-v t u s t v Pred T(s, t) , ( ) Google PageRank 10 / 23
(Bonacichi Centrality): . x = αA1 + βAx, x ,
1 1−z = 1 + z + z2 + z3 + . . . x = (I − βA)−1(αA1) = α ∞ k=0 βkAk+11, = α(A + βA2 + β2A3 + . . .)1. ⇒ i j , 1 , 2 , 3 , . . . β . α = 0, β = 1/λ , [aij ] . ( ) Google PageRank 11 / 23
, , i aij 1/ki , , PageRank . ,
Katz Hubbel . ⇒ , . ( ) Google PageRank 12 / 23
3. PageRank WWW , ( ) Google PageRank 13 /
23
r ← Pr, r P rv ← d v′∈Nv rv′
kv′ + 1 − d N , d ≈ 0.85, 1 − d ⇒ ( ) Google PageRank 14 / 23
Google i ri ⇔ πi , i πi = 1
, r ← d × Pr + (1 − d)/N π1 . . . πi . . . πj . . . πn T 1−d N . . . . . . . . . . . . . . . 1−d N . . . . . . . . . . . . . . . . . . . . . . . . . . . 1−d N . . . d ki + 1−d N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1−d N . . . 1−d N . . . . . . . . . . . . . . . . . . . . . . . . . . . 1−d N . . . . . . . . . . . . . . . 1−d N = π1 . . . πi . . . πj . . . πn T A.N.Langville and C.D.Meyer, Google’s Page Rank and Beyond, Princeton Univ. Press, 2006 ( ) Google PageRank 15 / 23
WWW ( ) Google PageRank 16 / 23
πi = d j Aij πj kout j + (1
− d)β j πj . π = dAD−1π + (1 − d)β1 → π = (1 − d)β(I − dAD−1)−11. Katz xi = α j Aij xi + β′. x = αAx + β′1 → x = β′(I − αA)−11 = β′ ∞ k=0 (αA)k1. A : x = 1 λ1 Ax = PageRank ! xi (t) = j Aij kj xj (t − 1), xi = ki j kj . x = AD−1x → (I − AD−1)x = (D − A)D−1x = 0, D−1x = 1. M.E.J.Newman, Networks -An Introduction-, OXFORD Univ. Press, 2010. ( ) Google PageRank 17 / 23
, , , , Gram-Schmidt Householder R Q QR Rayleigh-Ritz
Krylov x, Ax, A2x, A3x, . . . Arnoldi , Lanczos , Jacobi , , , , [ 2], 1994, ( ) Google PageRank 18 / 23
Google L.E.Page . , . WWW , . , WWW
, PC . ( ) Google PageRank 19 / 23
Indexing Hadoop https://enterprisezine.jp/dbonline/detail/4254 ( ) Google PageRank 20 / 23
MapReduce Map Key Value , Reduce Key http://www.dineshonjava.com/2014/11/mapreduce-flow-chart-sample-example.html ⇒ (
) Google PageRank 21 / 23
4. HITS J.M.Kleinberg HITS (Hyperlink-induced topic search) Authority: x ←
AT y ← AT Ax, x AT A Hub: y ← Ax ← AAT y. y AAT . A , AT . J.M.Kleinberg, Journal of the ACM 46, 1999 ( ) Google PageRank 22 / 23
5. i Li , . i Li N . ¯
L def = i Li /N . i ¯ L , ∆Li = Li − ¯ L . ∆Li < 0 . . , ( ) Google PageRank 23 / 23