anglc / nlp-sentimentclassification

project in school. Passive-Aggressive Algorithm, Language Modeling, and Winnow algorithm.

NLP SentimentClassification

blog content

Reference

2006 Comparative Experiments on Sentiment Classification for Online Product Reviews. Hang Cui, Vibhu Mittal, Mayur Datar here

Data Crawler

• 1000 positive and 1000 negative processed reviews
• 25 Movies So Bad They're Unmissable
• The 50 Worst Movies Ever
• Rotten Tomatoes > Movie > On Dvd & Streaming > Browse All

To Do

• add more features

Complete

• Simple Passive-Aggressive Algorithm here
• Simple Winnow algorithm here
• Simple Language Modeling here
• Passive-Aggressive Algorithm & Winnow algorithm adjust training here

Usage

• JavaSE-1.7
• Eclipse

If you compiler error with NOT FOUND MAIN CLASS on eclipse, find menu bar Project > Clean ... and run it to rebuild /bin.

Options

prompt> java -jar NLP-SentimentClassification.jar

Usage: java -jar NLP-SentimentClassification.jar [options]

Input Options

	-path <TRAINING_PATH>	training set folder path. default <TRAINING_PATH>="training_set".

	-tpath <TEST_PATH>		user test folder path. default <TEST_PATH>="user_test".

Processing Options	
	
	-n <NGRAM_MAX>			n-grams size. default <NGRAM_MAX>="3".

	-top <FEATURE_MAX>		pick the number of feature n-grams. default <FEATURE_MAX>="40000".

	-cross <CROSS_TIMES>	default <CROSS_TIMES>="5".

	-crosspart <RATIO>		pick <RATIO> : 5 = TRAINING : OTHER, default <RATIO>="5".

	-ittime <ITCOUNT> 		training model, default <ITCOUNT>="20"

	-oittime <ITCOUNT>		online training model, default <ITCOUNT>="20"

	-ui 					call UI interface.

Sample

Sample 1

java -jar NLP-SentimentClassification.jar -n 3 -top 40000 -path training_set -tpath user_test

Sample 2

java -jar NLP-SentimentClassification.jar -n 3 -top 10000 -path training_set2 -tpath user_test

output with markdown format, and classifier will output under path/output/. For example, ouotput/neg/Adaboost.txt.

<TRAINING_PATH> folder must give the file dictionary like this

.
├── extra		// support data, like dictionary, ban list, ...
├── neg 		// negative processed data 
└── pos 		// positive processed data

<TEST_PATH> folder must give the file dictionary like this

.
├── neg 		// test data which expected negative
└── pos 		// test data which expected positive

If you want to test unknown data, put them in user_test/neg or user_test/pos folder. Program will generate result of classifier to output/neg or output/pos folder.

Good Luck.