This folder creates the data necessary for simulation and user study using the Amazon 
kitchen review data (data source: https://www.cs.jhu.edu/~mdredze/datasets/sentiment/). 
This data has also been used in:
"Hernández-Lobato, et.al. (2015). Expectation propagation in linear regression models
with spike-and-slab priors. Machine Learning, 99(3), 437-487."


There are three scripts:
  extract_amazon_data.m: reads the review file, prepares the bag of words of the
  full data, and extracts the response values.
  
  prepare_amazon_data.m: This script removes keywords with less than 100 occurrence
  in the reviews (as suggested by the paper).
  
  learn_z_star.m: this script first performs cross validation on the data to learn good
  parameters for Spike-and-Slab linear regression. After that, it uses the best parameters
  to fit a model to the data. The output can be used as some kind of a ground truth of
  data. A visualization of the results can be found in the fig folder.

The folder "User Study" contains the relevance feedback of 10 participants on all the words
in the reviews. The feedbacks are 0:not relevant, 1:relevant, or -1:I don't know.