Extreme multi-label classification (or extreme multi-label learning, XML) is the task of matching an input with 0 or more labels (the most relevant labels) from an extremely large label set. The space of outputs is $2^L$, where $L$ is a large set. This is different from multi-class classification, where each instance has only one associated label. It has been used for recommendations and product search. Search and IR can also be formulated as XML.

• Wei et al 2022 - A Survey on Extreme Multi-label Learning
• Tarekegn et al 2024 - Deep Learning for Multi-Label Learning: A Comprehensive Survey

• Zhang et al 2018 - Deep Extreme Multi-label Learning
• You et al 2018 - AttentionXML: Label Tree-based Attention-Aware Deep Model for High-Performance Extreme Multi-Label Text Classification
• Applications
  • Chang et al 2021 - Extreme Multi-label Learning for Semantic Matching in Product Search

• Classification