# How to properly compare these two confusion matrix?

I have used Vader, a sentiment analysis tool for social media, on a database of movie reviews. These two confusion matrices differ in the vader.py algorithm, as the first one is from nltk:

The second one is deriving from Vader's original code on github and includes fixes to negation words, etc.

I was wondering how could I properly compare the two, as I'm not really able to read them. It seems there is not a big difference between them and I don't understand what could be the sources of the errors here.

• What do you want to compare?
– Dave
Sep 1 '20 at 14:34
• the two results given by the matrix, that is to say, two versions of the vader.py algorithm for sentiment classification (1 being very negative and 5 very positive).
– Anna
Sep 1 '20 at 15:21
• In what way do you want to compare them? You already know that they are slightly different.
– Dave
Sep 1 '20 at 15:24
• I just can't read the results
– Anna
Sep 1 '20 at 15:41
• What do you mean that you can't read the results?
– Dave
Sep 1 '20 at 15:44

First, about interpreting these confusion matrices: the sum of every row is 1, which implies that every value is a conditional probability p( predicted label | true label ), i.e. the probability of a given true label to be a particular predicted label. Example: the top left cell in both matrices is 0.01, which means that when the true label is 5 the probability that the system predicts label 1 is 1%.