Only using match-counts is imo a bit simplistic (don't get me wrong, simplistic can be good). Do you have any information like "how many times does this negativ

Only using match-counts is imo a bit simplistic (don't get me wrong, simplistic can be good). Do you have any information like "how many times does this negatively annotated word occur in a document"; then you can use a simple calculation (like cosine-similarity) to calculate a measure of matching with said case. Also, consider using bigrams (i.e. word-pairs) to do sentiment matching which will make matching more precise.