Extreme multi-label classification (XMC OR XMLC) is the approach used to assign the disclosed labels for the input from the extremely high number of possible labels. The main significance of extreme multi-label classification is understanding and learning the architectures and classifiers that automatically find each instance with the most relevant subset of labels from an exceedingly large set of labels. An increase in data size leads to data annotation with less accuracy. To overcome such issues in extreme multi-label classification, annotate high-quality data.
Several methods are developed for extreme multi-label classification, namely, one-vs-all, tree-based methods, label partitioning methods, embedding-based methods, probabilistic label tress methods, and flat neural methods. Extreme multi-label classification is also integrated with learning paradigms such as transfer, few, and zero-shot learning. Extreme multi-label classification applications are web directories, product categorization, indexing legal documents, categorizing medical examinations, image classification, question answering, advertising, and various applications in natural language processing.
Recently, Extreme multi-label classification has been applied in real-world tasks such as recommendation systems, document tagging, semantic matching, and advancements in XMC are deep transformer models and deep learning-based extreme multi-label text classification. Further research scope utilizes different deep learning models for extreme multi-label classification.