Papers › Hierarchy-aware Biased Bound Margin Loss Function for Hierarchical Text Classification

Hierarchy-aware Biased Bound Margin Loss Function for Hierarchical Text Classification

13 Aug 2024Findings of the Association for Computational Linguistics ACL 2024 8archive 2025-07-28

Gibaeg Kim, SangHun Im, Heung-Seon Oh

Hierarchical text classification (HTC) is a challenging problem with two key issues: utilizing structural information and mitigating label imbalance. Recently, the unit-based approach generating unit-based feature representations has outperformed the global approach focusing on a global feature representation. Nevertheless, unit-based models using BCE and ZLPR losses still face static thresholding and label imbalance challenges. Those challenges become more critical in large-scale hierarchies. This paper introduces a novel hierarchy-aware loss function for unit-based HTC models: Hierarchy-aware Biased Bound Margin (HBM) loss. HBM integrates learnable bounds, biases, and a margin to address static thresholding and mitigate label imbalance adaptively. Experimental results on benchmark datasets demonstrate the superior performance of HBM compared to competitive HTC models.

PaperPDFCode

Code

whitepurple/HBM-loss-for-HTC officialmentioned in paperpytorchMIT report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Hierarchical Multi-label ClassificationMulti-Label ClassificationText Classificationtext-classification

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Hierarchical Multi-label Classification EURLEX57K HiDEC+HBM Loss Macro F1 28.77±0.11 #1 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HiDEC+HBM Loss Micro F1 76.48±0.12 #1 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HPT+HBM Loss Macro F1 28.70±0.22 #2 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HPT+HBM Loss Micro F1 75.78±0.15 #2 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HPT Macro F1 28.46±0.26 #3 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HPT Micro F1 75.54±0.20 #3 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HiDEC Macro F1 27.91±0.11 #4 of 4 Archive leaderboard report
Hierarchical Multi-label Classification EURLEX57K HiDEC Micro F1 75.14±0.19 #4 of 4 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HiDEC+HBM Loss Macro F1 70.69±0.19 #1 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HiDEC+HBM Loss Micro F1 80.52±0.18 #1 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HPT+HBM Loss Macro F1 70.23±0.18 #2 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HPT+HBM Loss Micro F1 80.42±0.12 #2 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HiDEC Macro F1 69.80±0.24 #4 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HiDEC Micro F1 80.13±0.16 #4 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HPT Macro F1 69.69±0.49 #5 of 7 Archive leaderboard report
Hierarchical Multi-label Classification New York Times Annotated Corpus HPT Micro F1 80.04±0.23 #5 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HiDEC+HBM Loss Macro F1 71.47±0.20 #1 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HiDEC+HBM Loss Micro F1 87.81±0.09 #1 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HiDEC Macro F1 70.82±0.20 #2 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HiDEC Micro F1 87.70±0.12 #2 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HPT+HBM Loss Macro F1 70.55±0.13 #3 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HPT+HBM Loss Micro F1 87.82±0.06 #3 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HPT Macro F1 70.23±0.31 #4 of 7 Archive leaderboard report
Hierarchical Multi-label Classification RCV1-v2 HPT Micro F1 87.82±0.14 #4 of 7 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: HBM Loss

HBM LossZLPR Loss

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections