[peterjliu]

The code can be used to train on other data. All you really need is a collection of news articles. I think there are some free ones available.

This dataset was only used to benchmark against other published results. It was first proposed in https://arxiv.org/abs/1509.00685.