CSE 450 – Web Mining SeminarProfessor Brian D. Davison

Fall 2005

A Presentation on

When Experts Agree: Using Non-Affiliated Experts to Rank Popular Topics

K. Bharat & G. A. MihailaWWW10 Conference, May 2001, Hong Kong

byOsama Ahmed Khan

10/06/2005

ProblemProblem

Query on Popular Topic Content Analysis

SolutionSolution

Most Authoritative Pages

Technical Terms

Expert Recommendation Non-affiliation

Hilltop AlgorithmHilltop Algorithm

1. Expert Lookup Detecting Host Affiliation Expert Selection Expert Indexing

2. Target Ranking Computing Expert Score Computing Target Score

Detecting Host AffiliationDetecting Host Affiliation

Conditions Same first 3 octets of IP

127.0.0.1 127.0.0.15

Same rightmost non-generic token of hostname

www.ibm.com www.ibm.co.mx

Union-Find Algorithm

Expert SelectionExpert Selection

Retrieve all webpages with:

Out-degree > Threshold (k)

(e.g. k = 5)

Expert will have:

URLs pointing to k distinct non-affiliated hosts

Expert IndexingExpert Indexing

Inverted Index Mapping Keywords to Experts Key Phrases Match Positions

Computing Expert ScoreComputing Expert Score

Condition Atleast 1 URL with all query keywords

Expert Score: (S0, S1, S2)

Si = SUM{key phrases p with k-i query terms} * LevelScore(p) * FullnessFactor(p,q)

Expert_Score = 232 * S0 + 216 * S1 + S2

Computing Target ScoreComputing Target Score

Condition Atleast 2 non-affiliated experts

Target Score:

Edge_Score(E,T) = Expert_Score(E) *

SUM{query keywords w} * occ(k,T)

Target_Score = Sum{Edge_Score(E,T)}

EvaluationEvaluation1. Locating Specific Popular Targets

Evaluation Evaluation (Contd.)

2. Gathering Relevant Pages

ConclusionConclusion

Characteristics Popular Queries Expert Subset

Hilltop vs. PageRank Topic Distillation

Thank YouThank You