Bayesian naive classification ... not as effective

I have a database of 10 million products (name, description, brand, category) as a training dataset. I want to make an algorithm to classify about 10,000 products that have no category.

I made a small java program to train the algorithm using the Naive Bayes classifier, but when I enter my 10,000 products, I only get 30% correct answers ....

Is there a way to improve this?

Thank.