A Dynamic Centroid Text Classification Approach by Learning from Unlabeled Data
Cuicui Jiang, Dingju Zhu, Jiang Qingshan · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2013
AbstractThe centroid-based classification has proved to be a simple and yet efficient method for text classification.However, the performance of centroid-based classifier depends heavily on the quantity of labeled training set.It is easy and cheap to collect enormous unlabeled data from digital resources, while it is difficult and costly to label these data for training classifiers.To address this problem, we propose a dynamic centroid text classification approach which learns from unlabeled texts to construct dynamic centroids.The main idea of the approach is to take the unlabeled texts with high classifying confidence into consideration to adjust the centroids dynamically.Experiments on two public corpora have indicated the effectiveness of our text classification approach in the case of spare labeled training set.