基于Spark的多标签超网络集成学习

发布时间：2018-05-13 04:20

本文选题：多标签学习 + 超网络　；参考：《智能系统学报》2017年05期

【摘要】：近年来,多标签学习在图像识别和文本分类等多个领域得到了广泛关注,具有越来越重要的潜在应用价值。尽管多标签学习的发展日新月异,但仍然存在两个主要挑战,即如何利用标签间的相关性以及如何处理大规模的多标签数据。针对上述问题,基于MLHN算法,提出一种能有效利用标签相关性且能处理大数据集的基于Spark的多标签超网络集成算法SEI-MLHN。该算法首先引入代价敏感,使其适应不平衡数据集。其次,改良了超网络演化学习过程,并优化了损失函数,降低了算法时间复杂度。最后,进行了选择性集成,使其适应大规模数据集。在11个不同规模的数据集上进行实验,结果表明,该算法具有较好的分类性能,较低的时间复杂度且具备良好的处理大规模数据集的能力。
[Abstract]:In recent years, multi-label learning has received extensive attention in many fields, such as image recognition and text classification, and has more and more important potential application value. Despite the rapid development of multi-label learning, there are still two main challenges: how to use the correlation between tags and how to deal with large-scale multi-label data. In order to solve the above problems, based on the MLHN algorithm, a multi-label super-network integration algorithm based on Spark is proposed, which can effectively utilize tag correlation and process big data sets. This algorithm first introduces cost sensitivity to adapt to unbalanced data sets. Secondly, the evolutionary learning process of supernetwork is improved, and the loss function is optimized to reduce the time complexity of the algorithm. Finally, selective integration is carried out to adapt to large scale data sets. Experiments on 11 data sets of different sizes show that the algorithm has good classification performance, low time complexity and good ability to deal with large data sets.
【作者单位】：重庆邮电大学软件工程学院;重庆邮电大学计算智能重庆市重点实验室;
【基金】：重庆市基础与前沿研究计划项目(cstc2014jcyj A40001,cstc2014jcyj A40022) 重庆教委科学技术研究项目(自然科学类)(KJ1400436)
【分类号】：TP181

【相似文献】