deletions | additions       diff --git a/From_the_first_figure_you__.tex b/From_the_first_figure_you__.tex  index 7ecec08..d24ca9f 100644  --- a/From_the_first_figure_you__.tex  +++ b/From_the_first_figure_you__.tex 
...
From the first figure, you can also see that there are users "switching" class between the self-assessed and attributed datasets of a given trait. Are the self-assessed values unreliable? or the attributed ones? The following figure shows all the class "switchers". In terms of raw numbers, there are 42, 53, 39, 57, 52 switchers in the respective five traits. This is quite a big number (from a 1/4 to a 1/3 of the attributed dataset)  and certainly has an effect on training the classifiers.