Abstract:

The principle of relevancy for auxiliary problems (APs) selection is proposed to further improve the classification results in alternating structure optimization algorithm. Many kinds of APs are created by taking examples of Chinese syntactic chunking and Chinese semantic role labeling. Experiments demonstrate that the APs created can help target problems to obtain better classification results.

Key words: alternating structure optimization, structural learning, semisupervised learning, auxiliary problems, target problems, relevancy