Core Concepts
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.
Abstract
MatchXML addresses the Extreme Multi-label text Classification (XMC) problem by proposing an efficient text-label matching framework.
The framework includes training dense label embeddings using label2vec, building a Hierarchical Label Tree, and formulating multi-label text classification as a text-label matching problem.
Experimental results show that MatchXML achieves state-of-the-art accuracies on five out of six datasets and outperforms competing methods in training speed.
The framework involves fine-tuning a pre-trained encoder Transformer, extracting dense text representations, and utilizing static dense sentence embeddings.
MatchXML's approach combines sparse TF-IDF features, fine-tuned dense text representations, and static dense sentence features to train a linear ranker.
The paper is organized into sections covering related works, the proposed method, experimental results, and conclusions.
Stats
MatchXML는 다섯 개의 데이터셋 중 다섯 개에서 최신 정확도를 달성했습니다.
MatchXML은 여섯 개의 데이터셋에서 경쟁 방법보다 학습 속도가 우수했습니다.
Quotes
"MatchXML achieves state-of-the-art accuracies on five out of six datasets."
"MatchXML outperforms the competing methods on all the six datasets."