Extreme Classification

Extreme classification (XC) tackles the challenge of assigning data points to the most relevant labels from an extremely large set, often millions or more. Current research focuses on developing efficient and accurate training algorithms, often employing deep learning architectures with techniques like approximate nearest neighbor search and negative mining to manage the massive label space. These advancements are crucial for applications like search, recommendation systems, and question answering, where high-speed, accurate predictions over vast label sets are essential for improved performance and user experience. The field is also actively exploring methods to address issues like missing labels and imbalanced data, leveraging techniques such as knowledge infusion and graph regularization to enhance model robustness and accuracy.