Biomedical Named Entity Recognition through a Multi-Agent Meta-Learning Framework

Recognizing the biomedical named entity has become one of the most fundamental tasks in the biomedical knowledge discovery.A multi-Agent meta-learning framework is presented which incorporates multi-agent system and meta-learning method for the application of biomedical named entity recognition.In the base level,different learning agent is selected according to different classes of biomedical named entities.Through the communication between base learning agents,a base learning agent can get beneficial information from other related base learning agents and adjust its behavior so as to improve its learning performances.In the meta-level,the synthetic decision from the results of base learning agents is made by the meta-agent.Meta-agent and base learning agents are integrated with sensitive features set of corresponding named entity class according to local feature selection,which improve the system performance especially on minor classes.This approach effectively overcomes the disadvantages that only one model and global feature selection are used to identity all types of biomedical named entities.The experiments are carried on JNLPBA2004 test corpus with an F-sore of 77.5%.The results show that the brand-new multi-agent meta-learning framework is an effective approach and get promising results in biomedical named entity recognition.