Fig. 3

Comparison of active learning with informativity based, entropy-based, Gibbs error based, and conventional committee based method, and random selection against GRO’13 dataset. The learning curves for the TEES system under active learning (AL), using the Gibbs error based method (Gibbs Error), entropy based method(Entropy), conventional committee based method (ConventionalCommittee) and the proposed informativity method (Informativity), as well as the random selection (RS), when tested against the GRO’13 task dataset. The active learning method uses the unigram model