Set.add(...) has side effects by nature, which means that your filter method isn't "pure". In the given example, I don't see a concrete downside, but if you later wanted to make the stream parallel, you'd need to use a concurrent Set.
For comparison, here's a side-effect free solution using Guava's Multiset:
public static Set<String> findDuplicateWordsInText(String text) {
Multiset<String> wordMultiset =
ImmutableMultiset.copyOf(Splitter.on(' ').split(text));
return wordMultiset.entrySet()
.stream()
.filter(entry -> entry.getCount() > 1)
.map(Multiset.Entry::getElement)
.collect(Collectors.toSet());
}
It probably doesn't perform as well, but the intent should (hopefully) come across clearly.
