Multi-Agent Classifiers Fusion Strategy for Biomedical Named Entity Recognition

Recognizing the biomedical named entity has become one of the most fundamental tasks in the biomedical knowledge discovery. The multi-agent classifiers fusion approach proposed here was found to efficiently recognize biomedical named entity. We employ conditional random fields as our underlying classifier model and incorporate diverse set of features into system, the relativity between classifiers is utilized by using co-decision matrix to exchange decision information among classifiers. The experiments are carried on GENIA corpus with the best result of 77.88% F-sore. The multi-agent classifier fusion strategy proposed here is obviously superior to the individual classifier based method and more effective than the classifiers fusion approach of boosting and bagging.