Examples Initialization in Chinese Text Categorization

The generalization ability is a fundamental goal for a classifier in machine learning. The categorization results are influenced by the initialized examples in a nearest neighbor classifier. The generalization ability beyond the examples in training set is important in categorization. In this paper, we propose a particle swarm optimization with k means clustering algorithm for the nearest neighbor classifier's examples initialization to improve categorization performances. This classifier utilizes an iterative strategy, and the classifier's example initialization is based on clusters center and random examples. The new classifier is tested on a Chinese text corpus. The proposed classifier is compared against the nearest neighbor classifier with random initialization.