This repo contains code required to reproduce experiments in the paper titled:

"Detecting Useful Business Reviews using Stylometric Features"

It presents an approach to detect useful business reviews using stylometric features. We train n-gram and PCFG models to capture writing style and structural patterns and employ them in extracting useful features from review text. Using Yelp dataset, we demonstrate that useful reviews have distinct style of writing that can be utilized in detecting them and empirically evaluate the same.We report improved accuracy compared to using just lexical or bag of words features. We also find that domain of business reviews do not affect the importance of stylometric features.

Data to reproduce the experiments in the paper is shared at:

http://www.cs.utexas.edu/~vidhoon/data-review-exp.zip

Please refer to README inside for details. The README is also found in data/ folder in this repo to avoid downloading entire data. Last time I checked, size was 1GB (approx).

Components are coded in python, JAVA and some tools are written in shell/bash scripts.

Please cite this paper if you happen to re use code or preprocessed/ processed Yelp review data from this project:

Vidhoon Viswanathan, Raymond Mooney, Joydeep Ghosh, "Detecting useful business reviews using Stylometric Features" (submitted to Yelp Data contest)