Multi-label classification using hierarchical embedding

摘要: Multi-label learning deals with the classification of data multiple labels.Output space many labels is tackle by modeling inter-label correlations.Use parametrization and embedding have been prime focus.A piecewise-linear using maximum margin matrix factorization proposed.Our experimental analysis manifests superiority our proposed method. concerned class labels. This in contrast to traditional problem where every instance has a single label. (MLC) major research area machine community finds application several domains such as computer vision, mining text classification. Due exponential size output space, exploiting intrinsic information feature label spaces thrust recent years use focus MLC. Most existing methods learn linear entire training set hence, fail capture nonlinear spaces. To overcome this, we propose which uses model parametrization. We hypothesize that vectors conform similar are some sense. Combining above concepts, novel hierarchical method for multi-label Practical problems image annotation, categorization sentiment can be directly solved compare six well-known algorithms on twelve benchmark datasets. Our over state-of-art algorithm learning.