The Use of Unlabeled Data Versus Labeled Data for Stopping Active Learning for Text Classification

摘要: Annotation of training data is the major bottleneck in creation text classification systems. Active learning a commonly used technique to reduce amount one needs label. A crucial aspect active determining when stop labeling data. Three potential sources for informing are an additional labeled set data, unlabeled and that during process learning. To date, no has compared contrasted advantages disadvantages stopping methods based on these three information sources. We find use more effective than