Hybrid adaptation: integration of adaptive classification with adaptive context processing

Abstract

We propose a new method of adaptation in online handwritten character recognition. The method, called the "hybrid adaptation", integrates adaptive classification with adaptive context processing. Hybrid adaptation includes a mechanism that minimizes the negative effects of adaptation that might be caused by the integration. Online handwritten character recognition software with hybrid adaptation can be loaded on terminals having low memory capacity since our implementation of both adaptive classification and adaptive context processing does not require much memory. In our experiments, under the condition that all input strings had been input previously, the first-hit rate of hybrid adaptation was 99.0%, while that of non-adaptation was 93.3%, that of adaptive classification was 95.3% and that of adaptive context processing was 97.9%. In addition, we confirm that hybrid adaptation could enhance the level of satisfaction of the individual user.