MIT Press

Table 7:

Prediction results for trust averaged over 5 cross-validation splits. The number of features in each set is included in parentheses in the feature set column. We incrementally added each feature set and also included the individual performance of each feature set in parentheses in the precision/recall/macro-F1 columns.

Feature Sets	Precision	Recall	macro-F1
random	50.19	50.29	44.97
majority	40.27	50.00	44.61
data-driven (9538)	73.45	59.72	61.51
disfluency (5)	76.83	56.73	57.28
+ prosody (16)	78.90 (80.14)	60.52 (58.66)	62.74 (60.20)
+ sentiment (5)	78.75 (40.27)	61.19 (50.00)	63.61 (44.61)
+ uncertainty (3)	78.11 (40.27)	61.80 (50.00)	64.36 (44.61)
+ creativity (1)	77.94 (40.27)	61.90 (50.00)	64.48 (44.61)
+ complexity (16)	77.71 (55.86)	62.24 (50.31)	64.87 (45.77)
+ speaker traits (7)	77.55 (40.27)	63.34 (50.00)	66.10 (44.61)
all (9591)	74.26	60.34	62.37

Feature Sets	Precision	Recall	macro-F1
random	50.19	50.29	44.97
majority	40.27	50.00	44.61
data-driven (9538)	73.45	59.72	61.51
disfluency (5)	76.83	56.73	57.28
+ prosody (16)	78.90 (80.14)	60.52 (58.66)	62.74 (60.20)
+ sentiment (5)	78.75 (40.27)	61.19 (50.00)	63.61 (44.61)
+ uncertainty (3)	78.11 (40.27)	61.80 (50.00)	64.36 (44.61)
+ creativity (1)	77.94 (40.27)	61.90 (50.00)	64.48 (44.61)
+ complexity (16)	77.71 (55.86)	62.24 (50.31)	64.87 (45.77)
+ speaker traits (7)	77.55 (40.27)	63.34 (50.00)	66.10 (44.61)
all (9591)	74.26	60.34	62.37

View Large

This Feature Is Available To Subscribers Only