Combining labeled and unlabeled data with co-training
COLT' 98 Proceedings of the eleventh annual conference on Computational learning theory
An empirical study of smoothing techniques for language modeling
ACL '96 Proceedings of the 34th annual meeting on Association for Computational Linguistics
Head-Driven Statistical Models for Natural Language Parsing
Computational Linguistics
Coaxing confidences from an old friend: probabilistic classifications from transformation rule lists
EMNLP '00 Proceedings of the 2000 Joint SIGDAT conference on Empirical methods in natural language processing and very large corpora: held in conjunction with the 38th Annual Meeting of the Association for Computational Linguistics - Volume 13
Control models of natural language parsing
Control models of natural language parsing
Solving multiclass learning problems via error-correcting output codes
Journal of Artificial Intelligence Research
A machine learning approach to the identification of appositives
IBERAMIA-SBIA'06 Proceedings of the 2nd international joint conference, and Proceedings of the 10th Ibero-American Conference on AI 18th Brazilian conference on Advances in Artificial Intelligence
Semi-supervised learning for portuguese noun phrase extraction
PROPOR'06 Proceedings of the 7th international conference on Computational Processing of the Portuguese Language
| Hi-index | 0.00 |
The classifiers produced by the Transformation Based error-driven Learning (TBL) algorithm do not produce uncertainty measures by default. Nevertheless, there are situations like active and semi-supervised learning where the application requires both the sample's classification and the classification confidence. In this paper, we present a novel method which enables a TBL classifier to generate a probability distribution over the class labels. To assess the quality of this probability distribution, we carry out four experiments: cross entropy, perplexity, rejection curve and active learning. These experiments allow us to compare our method with another one proposed in the literature, the TBLDT. Our method, despite being simple and straightforward, outperforms TBLDT in all four experiments.