This is called a confusion matrix. It graphically displays the ability of the system to make correct predictions for each output classification.
The entries on the main diagonal are correct predictions.
The entries off the main diagonal tell you which labels were predicted incorrectly, and how frequently each label is predicted incorrectly and is confused with a different label.
For example, looking at the top row.
There were 7 examples where “heart” was the correct label.
In six cases, the prediction was correct.
In one case, the prediction was “sad face” instead of “heart”.