Effect of adding binary negated tokens

Table 2 shows a summary of the average accuracy with and without the addition of binary negated tokens. Only the results for Diabetes, CAD, Gout and Depression are shown. Diabetes and CAD were selected for their high number of 'N' judgments (12 for Diabetes and 16 for CAD out of the 611 records composing the first textual training batch). Gout and Depression were selected for their low presence of 'N' judgments (0 for both diseases in the first textual training batch). Results show an improvement in average accuracy for diseases with presence of 'N' judgments when negative tokens are added, and no hurt in the improvement for diseases with no presence of 'N' judgments.

Table 2: Summary of the average accuracy (with std. deviation in parenthesis) of 4-fold cross-validation runs with and without negative tokens added. Diabetes and CAD have a high presence of 'N' judgments (compared to other diseases) and Gout and Depression have no presence of 'N' judgments. These results were produced considering only the first training data batch for textual judgments (611 records).

Disease	No negative tokens added	Negative tokens added
Diabetes	0.877 (0.018)	0.887 (0.021)
CAD	0.804 (0.012)	0.811 (0.024)
Gout	0.945 (0.007)	0.946 (0.010)
Depression	0.876 (0.035)	0.874 (0.048)