Comparison with "generalized expectation criteria"

I want to compare my semi-supervised text classification algorithm with the GE-FL algorithm in “Learning from labeled features using generalized expectation criteria” (http://people.cs.umass.edu/~mccallum/papers/druck08sigir.pdf).

I want to evaluate both the algorithms on the basis of Macro F1 and human labeling efforts.

The GE-FL algorithm is semi-supervised. But the labeled features are discovered using complete labeled data set.
In my algorithm labeled features are discovered without using labeled data set.

How should we compare human efforts required in my algorithm to the efforts required in the GE-FL algorithm?

Consider following example:
For some data set, Macro F1 using GE-FL algorithm is 0.89 and the number of labeled features are 40.
For the same data set, Macro F1 using my algorithm is 0.88 and the number of labeled features are 30.
Which algorithm is performing better in this case?