The dataset version is a random ID that definitively identifies the data currently published in a particular dataset.

Each time you publish data for a dataset, the dataset version is set to a random ID. When you import new data (as a draft dataset), and go on to publish it, then the new data replaces the existing data and a new dataset version is generated.

If you own a dataset or have permission to reference it in a query, then you can check its dataset version using the Platform API. In particular, this lets you check whether the dataset you are referencing is the same as the one you referenced last time.