Predicting the impact of an image on Flickr
(Carnegie Mellon University IPTSE Winter School '14)
This project was carried out as part of the Carnegie Mellon University's IPTSE program, was mentored by Prof. Bhiksha Raj & Prof. Rita Singh from CMU and held at National Institute of Technology, Karnataka. I as part of a group of 5, worked on techniques to analyse, predict and exploit the impact of an Image from Flickr.
We extracted some interesting patterns from Images in Flickr, their seasonal variation. The nature of advertisements to other photos, the most popular tags etc. The project involved, correlation of image characteristics like color standard deviation, intensity with linguistic factors like sentiments etc.
We also utilised statistical techniques like Kernel Density Estimation on Geo Located data and tried applying ARMA models on the comments stream to predict the growth of comments on our images.
All scripts written by me for this project are available on github, here.
The results of our study were concluded in the form of a poster presented at the end of the 2 weeks long winter school. Which is available below