Given your statistics, it seems more than 90% of all articles are not popular. That means a 90% accuracy of your model is not impressive, since you get that by classifying all articles as non-popular. Did you check how many of the popular articles were correctly classified?