I've read this article about Cassandra Data modeling. But I didn't understand in which cases exactly Partially de-normalization is preferred? I've read some other articles which all of them use de-normalization (not partially).
Partial de-normalization means that you duplicate only some columns of the target table.
Full de-normalization means you duplicate all columns of the target table
In practise, it is recommended to de-normalize, IF POSSIBLE, only immutable data so that you don't need to care about their update. In this case, duplicating only immutable data == partial de-normalization.