Why does the SVM function in R make wrong predictions after constructing a DTM?

Question

My objective is to analyze the sentiment of 1000 labeled tweets. I have split the data into 80:20 training:test sets where the training set includes 50% of positive tweets and