link prediction and path analysis using markov chains (r. r. sarukkai) presentation by h.perrin,...

13
Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Upload: brooke-byrd

Post on 28-Mar-2015

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Link Prediction and Path Analysis using Markov Chains

(R. R. Sarukkai)

Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis

1

Page 2: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Link Prediction & Path Analysis

● Volume of pages makes efficient WWW navigation difficult

● Aim: To analyse users' navigation history to generate tools that increase navigational efficiency– ie. Predictive server prefetching

● Provides tools for other work previously done.

Page 3: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

The Author: Ramesh R. Sarukkai

● Researches Internet technologies

● Member of W3C committee

● Now works for Yahoo!

Page 4: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

The Theory Proposed

4

Page 5: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Background: Markov Chains

Page 6: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

The Theory

● (INCLUDING MATHS?)

Page 7: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

The System

Page 8: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Applications

8

Page 9: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

HTTP Request Prediction

● In server or proxy

● Allows pre-fetching of most likely pages first

● Connection latency is minimsed

● Server efficiency increases

Page 10: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Link Suggestion (Adaptive Web Navigation)● Link prediction used to offer links to users

based on previous navigation history

● Similar technique has been applied (ie. Amazon) – But not necessarily using Markov Chains

● Can be client or server side

Page 11: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Tour Generation

● Given a start URL, user guided along path of links

● Appropriate to user's interests

● Sequentially pick the next most popular link

● Non-cyclic

Page 12: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Hub/Authority Identification

● Kleinberg proposed Web 'Hub/Authorities' – Hub: Web site that is a good starting point for

finding identification

– Authority: Web site that contains useful information on a particular topic.

● Hub/Authority weighting given in Markov transitional weighting

Page 13: Link Prediction and Path Analysis using Markov Chains (R. R. Sarukkai) Presentation by H.Perrin, S.Jaffer, S.Lambert & W.Lewis 1

Experimental Results