Our investigation aims at constructing oblique decision stump forests to classify very large number of twitter messages (tweets). Twitter sentiment analysis is not a trivial task because tweets are short and getting generated at very fast rate. Supervised learning algorithms can thus be useful to automatically detect positive or negative sentiments. The pre-processing step performs the cleaning tasks and the representation of tweets using the bag-of-words model (BoW). And then we propose oblique decision stump forests based on the linear support vector machines (SVM) that is suitable for classifying large amounts of high dimensional datapoints. The experimental results on twittersentiment.appspot.com corpora (with $1,600,000$ tweets) show that our oblique decision stump forests are efficient compared to baseline algorithms.