Accuracy on tweets is much lower than reported in paper

[sourface94]

HWhen reproducing the experiments using this code when trained on the ~10000 row dataset and testing against a users tweets, each user I test always get the ENFJ personality with all scores +/- 0.15 from:
Score for O is: 4.3
Score for C is: 3.4
Score for E is: 3.4
Score for A is: 3.6
Score for N is: 2.6
Jungian type is ENFJ

Do you have any idea why this may be?

I am using a mypersonality dataset with 9918 rows (excluding header) compared to your dataset with 9913 rows, however I think we may be using the same dataset and you removed a few rows?