Abstract: Multi-label classification methods have been applied in many real-world fields. To the problem of label-relevance in multi-label classification, this paper proposes a novel multi-label classification method by automatically acquiring label relevant information. The method introduces a probabilistic model to solve the established optimization sub-problem by using alternative maximization algorithm. It also gives the inference. This model can acquire label relevant information, and it can achieve better multi-label classification results. The experimental results on four real multi-label datasets show that the proposed method can achieve higher classification and prediction evaluation values and several other measure index values than the existing multi-label classification methods.

Key words: multi-label classification, label relevant information, probabilistic model, alternative maximization algorithm

摘要： 多标签分类已在很多领域得到了实际的应用。针对多标签分类中存在标签相关性问题，提出一种获取标签相关信息的多标签分类新方法，记为LRI_MLC。该方法主要是通过引入一个概率模型来实现，即对建立的最优化子问题采用交替最大化法进行求解，并给出了求解推到过程，自动地获得标签的相关信息，以达到较好的多标签分类效果。在四个多标签数据集上的实验结果表明，提出的方法得到了较好的分类预测评价值以及其他几种衡量指标值，优于现有经典的多标签分类方法。

关键词: 多标签分类, 标签关联信息, 概率模型, 交替最大化算法