I've already tokenized the field, so, for example, 'kw1' appears 10 times across all document, 'kw2' 20 time, and so on and so force.

If I check index with Luke, it will be looking like:



I need get top N highest ranking terms. It is something like Google trend.

Any idea?

Lucene is perfect for this kind of use cases, but you won't issue a standard Query nor load entities right?
The terms with associated frequencies are stored in the index, open an IndexReader and use "low level" Lucene API to read the values.