In order to extract effective features from text, we propose a novel feature selection method based on improved binary particle swarm optimizer. The improved binary particle swarm optimizer iterates according to round angle, local round factor and global round factor, search best values of fitness function, and then select those feature with weight 1, and ignore those features with weight 0. The experimental results show that the method not only cuts down computing cost, and is helpful to improve precision of bio-entity recognition.