I train a classifier on a set of movie reviews to try to create a model for sentiment. Reviews are tokenized using TFIDF and then the respective TFIDF scores are used to train an SVM classifier. This classifer now knows a very basic understanding of what words correspond to positive and negative emotions/events. From here, incoming tweets are tokenized and the generated model is used to predict the sentiment of the tweet.