G ROUP PROXIMITY MEASURE FOR RECOMMENDING GROUPS IN ONLINE SOCIAL NETWORKS Barna Saha and Lise Getoor University of Maryland SNA-KDD Workshop ‘08 Presented.

GROUP PROXIMITY MEASURE FOR RECOMMENDING GROUPS IN ONLINE SOCIAL NETWORKSBarna Saha and Lise Getoor

University of Maryland

SNA-KDD Workshop ‘08

Presented by

Sai Moturu

Oct 17

OVERVIEW

Setting: Communities in Online Social Networks

Goal: Recommending groups/communities to users

Problem: Defining proximity between communities

Approach: Group Proximity Measure

Experiments: Flickr, Live Journal, You Tube

ESCAPE PROBABILITY

Ei,j – Escape probability from i to j – probability that a random walk from node i will visit node j before visiting i

Vk(i,j) – Probability that a random walk from node k will visit node j before visiting node i

Computed using the Fast algorithm by Tong et al.

APPROACH OUTLINE

Let Gi and Gj be two groups Ci/Cj represents the core and Oi/Oj represents the

outliers Find CORE

Find Ci & Cj Obtain Concise Graph

Shrink Ci & Cj into two vertices Vi & Vj Remove self loop and replace parallel edges with a single

edge and representative weight Call the concise graph G’

Compute Escape Probability in G’

FINDING CORE

Degree Centrality For a node, its degree in the group is the number of

members of the group it is linked to Pick all members with a degree above a certain threshold

Subgraph Pick the subgraph within a group that has maximum ratio of

edges/vertices

OBTAIN CONCISE GRAPH

PREDICTING FUTURE GROWTH

Link Cardinality Estimation Group Proximity Measure Number of links in between Product of the size of the two groups

Classification

GROUP RECOMMENDATION MODELS

RESULTS

RESULTS

CONTRIBUTIONS

New link-base proximity measure for groups in online social networks

Using proximity measure and structural properties to predict number of new links that will develop between two groups

New recommendation system based on group proximity and history of user’s group membership

THANK YOU