RK-A

Today, it is roughly proportional to the number of levels in the section hierarchy that have nouns. In a little more detail, it looks like this:

if (numberOfNounsInKpSentence 0 or NumberOfNounsInSimilarDocHeader 0 )
then penalty = 0.015
else:
-matched_words = check_context_match(numberOfNounsInKpSentence, NumberOfNounsInSimilarDocHeader );
-if matched_words are under 5k:
--if words are under 2k then 0.03 else: 0.015 penalty
--elif matched_words are empty(means no context matched):
---if sim_doc_header/parent_header_list length is 3 and in comparing first level with second level hierarchy Noun words are all same
----then 0.03 penalty
---else:
----penalty by decrease_score_by_context_match();
--else:
---no context match penalty

where decrease_score_by_context_match() consist of
penalty = 0.03 + (len(parent_header_list) - 1) * 0.01
and also penalties for some specific few cases.

Going forward, we would like to change the penalty to be proportional to numberOfNounsInKpSentence and NumberOfNounsInSimilarDocHeader.