Skip to content

Latest commit

 

History

History
29 lines (21 loc) · 1.36 KB

README.md

File metadata and controls

29 lines (21 loc) · 1.36 KB
Raw

Time Sensitive Influence Maximisation

Project for the CS229 (machine learning) course at KAUST.

Abstract

In a diffusion network, the estimated likelihood of influence of one node on another depends on the transmission rate of the influencer node. Existing approaches learn the transmission rates and influence likelihoods from data, but consider the transmission rate to be time- invariant. This project proposes a modification by modeling the transmission rate as a distribution over time; specifically, over the 24 hours of a day. New influence estimation and maximisation algorithms are proposed in this model, and implemented as modifications of ConTinEst and InfluMax, respectively.

Implementation

This project modifies the influence propogation model to consider transmission rates of nodes that not static, but a distribution over the 24 hours of a day. This approach is then composed of two parts:

  1. Influence estimation, built on ConTinEst.
  2. Influence maximisation, built on InfluMax.

Data

Both ConTinEst and InfluMax provide toy data and synthetic data generators. I will evaluate my performance on these toy/synthetic data sets, and probably on a real dataset from Twitter if time permits.