Preprocessing(clean tweets), create model to predict sentiment-label(positive, negative or neutral) for tweets and model accuracy checking.

• Vasiliki Koumarela
• Lazaros Avgeridis

• Python version 3.6
• NLTK
• Numpy

jupyter notebook
pip install --user -U nltk
pip install --user -U numpy
pip install vaderSentiment

• Cleaning the data(process tweets from train.tsv and test.tsv using: Tokenization, StopWord filtering, Stemming)
• Make workclouds and matplots for the data
• Vectorization(using: BAG-OF-WORDS & TF-IDF)
• TSNE model(Word2vec)
• Classification: KNN , SVM
• Check the accuracy from label predictions

Use f1_score to calculate the success rate (classification labels with the official labels of test tweets SemEval2017_task4_subtaskA_test_english_gold.txt)

Model Accuracy: 0.59 success