基于元学习策略的分类器融合方法及应用

提出了基于元学习策略的分类器融合的新模型,使用了两类元学习策略将4种分类算法即Generalized Winnow算法、支持向量机算法、条件随机域算法和最大熵算法进行融合,并根据具体领域的应用任务和分类器特点选择了有效特征信息,在面向生物医学文本命名实体识别的应用中取得了较高识别精度。实验结果表明基于元学习策略的分类器融合方法明显优于单分类器方法,并且也优于基于判别规则的分类器融合方法。

Meta-learning based classifier ensemble strategy and its application

A novel meta-learning based classifier ensemble model was presented. Four classifiers i.e. Generalized Winnow, support vector machine, conditional random fields, and maximum entropy were combined using two different meta-leaming strategies. Various evidential features specified for the application of biomedical named entity recognition were incorporated into the system to help improve recognition performance. Experimental results show that the classifier ensemble strategy based on meta-learning is obviously superior to the individual classifier based method and superior to the arbitration rule based ensemble method.