topic-sensitive pagerank
DESCRIPTION
Topic-Sensitive PageRank. Taher H. Haveliwala Stanford University Presentation by Na Dai. The frame of system using topic-sensitive PageRank. PageRank. Rank is a n-dimension column vector of PageRank values.(i.e. Rank = (Rank(1), Rank(2),…, Rank(n)) T Motivation: irreducible & aperiodic - PowerPoint PPT PresentationTRANSCRIPT
Topic-Sensitive PageRank
Taher H. Haveliwala
Stanford University
Presentation by Na Dai
The frame of system using topic-sensitive PageRank
PageRank
• Rank is a n-dimension column vector of PageRank values.(i.e. Rank = (Rank(1), Rank(2),…, Rank(n))T
• Motivation: irreducible & aperiodic– Dangling node (Matrix D)
– Damp factor α(Matrix E)
Topic-Sensitive PageRank (1)
• w (w1, w2,…,w16): a normalized vector with length 1• wi = Pr(ci|q)
p
v1 v2 … … v16
w1 w2 w16
v1i=1/|T1| for i∈T10 else
v2i=1/|T2| for i∈T20 else
v16i=1/|T16| for i∈T160 else
α, M, D, Rank(i)
Rank(i+1)
Topic-Sensitive PageRank (2)
p
v1 v2
… …
v16
v1i=1/|T1| for i∈T10 else
v2i=1/|T2| for i∈T20 else
v16i=1/|T16| for i∈T160 else
α, M, D, Rank2(i)
Rank2(i+1)
p pα, M, D, Rank1(i)
Rank1(i+1)
α, M, D, Rank16(i)
Rank16(i+1)
Rank
w1 w2 w16
Effect of ODP-Biasing (1)
Effect of ODP-Biasing (2)
Effect of ODP-Biasing (3)
Query-sensitive Scoring
Query-sensitive Scoring
Future Work
• Investigate the best basis topics– Topic granularity– Topics that are deeper in hierarchy
• vj: resistant to adversarial ODP editors