MatchXML: Efficient Text-label Matching Framework for Extreme Multi-label Text Classification
핵심 개념
MatchXML proposes an efficient text-label matching framework for extreme multi-label text classification, achieving state-of-the-art accuracies and outperforming competing methods in training speed.
초록
Introduction to eXtreme Multi-label text Classification (XMC)
XMC aims to annotate input text with relevant labels from a large label set.
Proposed Method: MatchXML
Utilizes label2vec to train semantic dense label embeddings.
Constructs Hierarchical Label Tree by clustering dense label embeddings.
Formulates multi-label text classification as a text-label matching problem.
Experimental Results
Achieves state-of-the-art accuracies on five out of six datasets.
Outperforms competing methods in training speed.
MatchXML
통계
"Our experiments demonstrate that the dense label embeddings can capture the semantic label relationships and generate improved HLTs compared to the sparse label embeddings."
"MatchXML achieves the state-of-the-art accuracies on five out of six datasets."
인용구
"We propose MatchXML, an efficient text-label matching framework for XMC."
"Experimental results demonstrate that MatchXML achieves the state-of-the-art accuracies on five out of six datasets."