shivgarg / twitter-sentiment Goto Github PK

#Twiiter Sentiment Analysis

Twiiter Sentiment analysis tool written in Python using Scikit for running the machine learning models. The sentiment considered here is binary (0 or 4). 0 meaning negative sentiment of tweet and 4 meaning positive sentiment of tweet.

sklearn package should be installed and path variable should be confgiured correctly so that the

Repo Structure

1. add_features.py :- It parses the raw data file and cleans the data and add feaures to each tweet. It outputs a clean file which can be used to train a scikit model.
2. train.py :- This uses the cleaned up file and trains the scikit model. The experiments with various models can be run in this file by changing the models. It dumps the model in the models folder alongwith the vectorizer which acts the dictionary for the model.
3. test.py :- This file takes unlabelled data and outputs sentiment labels.
4. data:- folder contained the orginal data set plus the cleaned up datset.
5. model :- Contains the learnt model.
6. vector.zip :- Containes the vectorizer outputed by the tf-idf vectorizewr of scikit.
7. misc :- Some miscellaneous files used for developing the features.

##Basic Features Many of the below features were tried to increase the accuracy of the baseline classifier. You can try these in your experiments.

• Try different classifiers like SVMs, random forests or even multilayer perceptrons.
• If you use logistic regression try playing with regularizers - try L1 instead of L2.
• Try lemmatizing
• Get rid of stop words and highly infrequent words.
• Use tfidf-based weighting.
• Use existing sentiment resources discussed . Examples: SentiWordnet or General Inquirer.
• Try bigrams (in addition to unigrams).
• Do tweet normalization on words where letters are repeated for intensity, e.g., haaaaaaate and loooooooove. You can even do more normalization using some internet slang dictionary.
• Use pos-tag each word and use the tagged word as a feature instead of the original word. You can use presence of capitalization or all caps as features. You could use this code for POS tagging of tweets: https://github.com/aritter/twitter_nlp

Experiments

The accuracies with various classifier that were tried:-

• Linear SVM : 77.8%
• Random Forest : 72.8%
• Multinomial Naive Bayes: 76.5%
• Bernoulli Naive Bayes: 76.4%
• Ridge Regression: 77.5%
• Perceptron : 70.5%
• Passive Aggressive Classifier : 70.1%

Running the Code.

1. python add_features.py "tweet_file". It would output a "tweet_file"_twt file.
2. python train.py "cleaned tweet file" "label file for each tweet".
3. python test.py "cleaned tweet file" "output file name". It gives out the sentiment labels in output file.