I won't post the code here, but here's how I solved it:

1. Setup (spliting words in spaces and special characters)
2. Create a dictionary counting the ammount of words for each text
3. Create a global dictionary for the total ammount of words
4. Normalize each local dictionary, considering the global frequency
5. Considering each word as a vector, calculate cosine distance between every text from setA versus setB.
6. Create similarity matrix
7. Find the permutation matrix to maximize PxS

Got 100% with this approach