论文部分内容阅读
在多标记学习中,发现与利用各标记之间的依赖关系能提高学习算法的性能.文中基于分类器链模型提出一种针对性的多标记分类算法.该算法首先量化标记间的依赖程度,并构建标记之间明确的树型依赖结构,从而可减弱分类器链算法中依赖关系的随机性,并将线性依赖关系泛化成树型依赖关系.为充分利用标记间的相互依赖关系,文中采用集成学习技术进一步学习并集成多个不同的标记树型依赖结构.实验结果表明,同分类器链等算法相比,该算法经过集成学习后有更好的分类性能,其能更有效地学习标记间的依赖关系.
In multi-mark learning, finding and utilizing the dependencies between each mark can improve the performance of the learning algorithm.In this paper, we propose a multi-mark classification algorithm based on the classifier chain model, which first quantifies the degree of dependence among markup, And construct a clear tree-dependent structure between tags, which can weaken the randomness of dependencies in the classifier chain algorithm and generalize the linear dependencies into tree-dependent relationships.In order to make full use of interdependencies between tags, The integrated learning technique further learns and integrates a number of different tag tree-dependent structures.The experimental results show that the algorithm has better classification performance after integrated learning than the classifier chain algorithm, which can learn the tag more effectively Dependency between.