Hey, here are the answers to your questions. Yes, I created the dataset manually; I agree that it’s not extensive, which is why the accuracy is reported on a completely different dataset and not the test dataset, just to be fair. This article tries to present an easy beginner friendly approach, which aims to only be the starting point for further exploration. W larger dataset would render better and more seasoned results, however I am not aware of any such specific datasets, creating one would be a manual effort.

I hope that answers your question. Thanks!