Table 5

Results for the causal effect of Adjectives on sentiment classification on Reviews. We compare TReATE(O,CF) to the ground truth ATEgt(O) and the baseline CONEXP(O). Confidence intervals ([CI]), computed using the standard deviations of ITEgt(O), TReITE(O,CF), and CONEXP, are provided in square brackets.

ExperimentATEgt(O)TReATE(O,CF)CONEXP(O)
Balanced 0.397 0.385 0.01
[CI[0.377,0.417] [0.381,0.389] [0,0.044]
Gentle 0.376 0.351 0.094
[CI[0.361,0.392] [0.347,0.355] [0.061,0.127]
Aggressive 0.634 0.603 0.126
[CI[0.613,0.655] [0.588,0.618] [0.095,0.158]
ExperimentATEgt(O)TReATE(O,CF)CONEXP(O)
Balanced 0.397 0.385 0.01
[CI[0.377,0.417] [0.381,0.389] [0,0.044]
Gentle 0.376 0.351 0.094
[CI[0.361,0.392] [0.347,0.355] [0.061,0.127]
Aggressive 0.634 0.603 0.126
[CI[0.613,0.655] [0.588,0.618] [0.095,0.158]
Close Modal