Multi-Agent Classifiers Fusion Strategy for Biomedical Named Entity Recognition

AbstractRecognizing the biomedical named entity hasbecome one of the most fundamental tasks in thebiomedical knowledge discovery. The multi-agentclassifiers fusion approach proposed here was found toefficiently recognize biomedical named entity. Weemploy Conditional Random Fields as our underlyingclassifier model and incorporate diverse set offeatures into system, the relativity between classifiersis utilized by using co-decision matrix to exchangedecision information among classifiers. Theexperiments are carried on GENIA corpus with thebest result of 77.88% F-sore. The multi-agentclassifier fusion strategy proposed here is obviouslysuperior to the individual classifier based methodand more effective than the classifiers fusionapproach of boosting and bagging5 Halama