| 摘要: |
| 高阶多标记学习算法能够挖掘所有类别标记之间的关系或标记子集的关系，但在处理错误传播和冗余或错误的类别依赖关系这两个问题上存在弊端。针对此类问题，在链式分类器的基础上，提出稀疏链式多标记学习算法（Sparse Classifier Chains，SCC）；为了验证所提出算法的有效性，将其与5种多标记学习算法进行对比，选取8个评价指标来评估算法的性能，在12个标准数据集上进行了实验验证，并利用秩和检验方法来分析所有对比算法之间的相对性能；实验结果表明：稀疏链式多标记学习算法优于所有对比算法，显著优于部分算法，并具有较强的泛化性能。 |
| 关键词: 多标记学习 稀疏链式分类器 高阶标记相关性 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| arse Classifier Chains for Multi Label Classification. |
|
WANG Can， HUANG Jun， QIN Feng
|
| Abstract: |
| Higher order multi label learning algorithm can mine the relationships or subsets of all class labels， however， it has some drawbacks in dealing with the problems of error propagation and redundant or erroneous class dependencies. For such problems， a new multi label classifier SCC， i.e.， Sparse Classifier Chains， was proposed based on classier chains by introducing sparse learning. In order to verify the effectiveness of the proposed algorithm， we compared this with other 5 kinds of multi label learning algorithms and selected 8 evaluation indexes to evaluate the performance of them. The experimental verification which was carried out on 12 standard data set lastly analyzed by using the rank sum test method to compare the relative performance . The experimental results showed that SCC was superior to all the contrast algorithms， significantly superior to some algorithms； furthermore， it had a strong generalization performance. |
| Key words: multi label learning sparse classifier chains high order label correlation |
