How exactly did you generate the features column for the sentiment classification task?
In you paper, you say you used word2vec to generate a feature vector for each word. How do go from feature vector for each word to a feature vector for each entire review? Did you just average the word feature vectors?