Upon closer examination of the definition of Cosine Similarity he realized that my compare subroutine included too many cosine functions. After editing the subroutine and duplicating a document in the corpus, a correct value of 1000 is returned for exactly similar documents. (Actually, it sometimes returns scores of 999 which I’m going to chalk up to rounding errors.)

“Thank you, Allen.”

Now a new problem presents itself. Specifically, the similarity scores for all the other documents are upside down:

  Comparison: scores closer to 1000 approach similarity

      d1    d2   d3   d4   d5   d6

  d1   -   396  459  538  541  320
  d2   -    -   478  247  334  240
  d3   -    -    -   312  304  265
  d4   -    -    -    -   694  438
  d5   -    -    -    -    -   367
  d6   -    -    -    -    -    - 

  d1 = aristotle.txt
  d2 = hegel.txt
  d3 = kant.txt
  d4 = librarianship.txt
  d5 = mississippi.txt
  d6 = plato.txt

Previously, hegel.txt (d2) and plato.txt (d6) where considered very similar, but now they are almost opposites. Something is still not correct, and I sincerely have no idea where to begin looking for a solution.

I have updated the downloadable scripts, but as far as the compare subroutine goes, they are still not perfect (broken).