Learning Social Influence from Past Data

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Presentation

Title: "Learning Social Influence from Past Data"

By

Edbert Eddie PUSPITO

Abstract

When a user performs an action in social networking websites, friends of 
this user may be influenced to perform the same or similar action. This 
phenomenon is called a social influence. Recent studies on data mining and 
machine learning try to model the phenomenon in social network websites and 
tried to answer the following question: if a group of Facebook users likes 
a photo, how many likes of this photo are there at the end?

This project proposes a new approach to calculate the probabilities of an 
influence of one social networking user over another user. Past research 
studies calculated the probabilities in the perspective of the user who 
performs the action. We shifted the paradigm and tried to calculate the 
probabilities in the perspective of the person who observed the actions.

We verified our ideas and techniques using the last.fm dataset consisting 
of a social graph with 83K nodes and 2M edges, together with an action log 
consisting of 110M actions. Experimental results showed that the new model 
performed better than previous models in correctly predicting the users 
that will perform the action.

Date:                   Monday, 27 April 2015

Time:                   5:20 - 6:00pm

Venue:                  Room 5505
                        Lifts 25/26

Committee Members:      Dr. Raymond Wong (Supervisor)
                        Dr. Pan Hui (Reader)