Identifying the gender of Twitter users using style-based features and the content-based features in combination with a Support Vector Machine (SVM) classifier.
This file contains the extraction of the content-based features and the experiments using the SVM classifier. We examined four style-based features: counts of mentions, hashTags, Urls and punctuation signs features.
This file contains the extraction of the structural-based features and the experiments using the SVM classifier. We examined tf-idf of n-grams, bag of words n-grams, part of speech n-grams.
This file contains the appropriate preprocessing steps that followed for the content-based features (removing twitter bias such as @mentions, hashtags, URLs, ‘@username’ and stripped of the character ‘#’ from the hashtags). There was no preprocessing done for style-based features.
SVM classifier with a linear kernel.
The accuracies of different C parameters for each feature.
Create Part of speech files for each user using the Stanford code.
This folder contains the Part of speech files for each user.
This folder contains the preprocessing files for each user.
This folder contains the tweets collected from the PAN-AP-2015 corpus from Twitter in English.
These files contain the counts for each style-based feature and the true labels for each user.
Extract counts of mentions, hashTags, Urls and punctuation signs features:
python structural.py --extract
Run the experiments using the style-based features:
python structural.py
Run the experiments using the bag of unigrams, bigrams, trigrams:
python content.py --bag uni
python content.py --bag bi
python content.py --bag tri
python content.py --bag all
For running the experiments using the tfidf n-grams: --tfidf uni, --tfidf bi, --tfidf tri, --tfidf all.
For running the experiments using the pos n-grams: --pos uni, --pos bi, --pos tri, --pos all.
For plotting the C parameters and the accuracy for each feature: python content.py --plot