• Implemented Doc2Vec to vectorize scrapped stock market news titles and classify if the articles encouraged buying, selling or holding

• Used the SMOTE data balancing method to have a better distribution of training data classes

• Experimented with a variety of models in TensorFlow to find the best classifier: SVM, neural network, decision tree, random forest, KNN

• Obtained an accuracy of 92 % on 10-fold cross validation and 56% on unseen data