Bring a coded matrix. Get back the structure it supports.

What you need — one CSV
A matrix of documents by concepts. One row per document, one column per concept, and a score or null in every cell:
corres does not know how you arrived at the number in a cell. It might be how often something occurs, a position on a scale like the one below, or a measure of your own. Because it cannot know, it does not guess: you make two choices, and the server applies them.
First, whether to normalise. Normalising divides each cell by the largest value in its own document, so every document is read as a percentage of its own maximum. That helps when documents differ in length or in how heavily they were coded, and hurts when your numbers already sit on a fixed scale that means the same thing in every document.
Second, where “important” starts. Half the largest value suits a fixed scale — on a 0–3 scale that is 1.5, so the top half counts. Half the average of the non-zero values suits counts, which have no ceiling to take half of. Or set your own number. Whichever you choose is shown against the data before you commit, and reported back with the result.
The scale below is one common way of coding, not a requirement:

0	Absent	the concept does not appear
1	Mentioned	brief reference only
2	Important	discussed and emphasised
3	Central	a core focus of the document

Minimum: a matrix should have siginificantly more documents than terms, and a significant range of values. Very small, square, or even matrices will be declined, with reasons provided. Bring a real data set that produces significant results using other methods.