Table 3 Average recall, over 30 classes, for several precision thresholds on the test set, comparing classifiers trained solely on LDA (1000 topics using text features), Append (LDA, audio, visual), fusion of LDA with Append on audio-visual features (LDA+Append), and fusion of all three feature types (LDA+audio+visual). While LDA feature alone perform very well, fusion, in particular of audio, video, and LDA features, does best

From: On using nearly-independent feature families for high precision and confidence

  Prec.
99 % 95 % 90 % Max F1
LDA 0.58 0.79 0.85 0.94
Append 0.65 0.86 0.91 0.93
LDA+Append 0.73 0.85 0.92 0.95
LDA+audio+visual 0.76 0.88 0.94 0.95