##Comparison of open source NLP tools for sentiment analysis on Yelp Dataset Challenge
For the results check the blogpost here.
This is a project for the Data Mining course taught in the undergraduate programme of the Computer Science & Engineering department of the Univerisity of Ioannina during the fall semester 2014/2015. For details checkout the handout.
###Goal
The objective of this project is to apply various sentiment analysis techniques(NLP) on the restaurant reviews and assess whether they can correctly identify the reviews as positive or negative.
Yelp has released an anonymized part of their stored data to the public. This was accompanied by [a challenge] (http://www.yelp.com/dataset_challenge) with various awards in order to incentivize research and generate insights for the use of the data.
Here follows a brief explanation of the dataset, from their website.
The Challenge Dataset includes data from Phoenix, Las Vegas, Madison, Waterloo and Edinburgh:
* 42,153 businesses
* 320,002 business attributes
* 31,617 check-in sets
* 252,898 users
* 955,999 edge social graph
* 403,210 tips
* 1,125,458 reviews
This project is not a participation in the challenge.
Three different open source tools have been used and assessed:
- Training and testing using the dataset on a Naive Bayes classifier
- Training with generic lexicons such as WordNet and SentiWordNet
- Stanford's CoreNLP
The code can be found in github. A brief explanation on how to run the code.
Make sure you have the following libraries installed before running the code.
Also you must have installed the stopword corpora of NLTK. To bring up the NLTK downloader, run the following in a python console.
import nltk
nltk.download()
This must be done before runing any of the classifiers below.
You need to provide the category of the businesses and the quantity of samples for each review class (pos/neg). The script creates two json files one for each class.
python extract_reviews.py 'Restaurants' 1000
You need to provide the category, the number of samples for each class and the number of folds for the k-fold cross validation. It trains one classifier for each feature extraction filter (single words, stopwords removal, bigrams, bigrams & stopwords removal) and prints the overall accuracy.
python run_bayes 'Restaurants' 1000 5
You only need to provided the category and the number of samples.
python run_sentiwordnet 'Restaurants' 1000
You need to add to java's classpath the path to the simple-json and corenlp jar.
First compile the corenlp.java file to create the class.
javac -cp ".:PATH_TO_SIMPLE_JSON/*:PATH_TO_CORENLP/*" -d . corenlp.java
Then run the compiled class providing the category and number of samples.
java -cp ".:PATH_TO_SIMPLE_JSON/*:PATH_TO_CORENLP/*" corenlp restaurants 1000