Extreme Multi-label Completion for Semantic Document Labelling with Taxonomy-Aware Parallel Learning

Christophe Broillet; Julien Audiffren; Ljiljana Dolamic; Philippe Cudr\'e-Mauroux

REVIEW 5 major objections 5 minor 50 references

Extreme Multi-label Completion for Semantic Document Labelling with Taxonomy-Aware Parallel Learning

T0 review · 5 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read TAMLEC claims a taxonomy-splitting transformer beats state-of-the-art extreme multi-label completion on EURLex, MAG-CS and PubMed.

desk verdict Genuine extension of HECTOR with a coherent architecture, but the strong empirical claims rest on a clean-prefix protocol that may not transfer to messier real incomplete labels. read the letter →

arxiv 2412.13809 v1 pith:PP7NIQPS submitted 2024-12-18 cs.LG

Julien Audiffren , Christophe Broillet , Ljiljana Dolamic , Philippe Cudré-Mauroux This is my paper · ORCID

classification cs.LG

keywords extrememulti-labelclassificationlabelcompletiontaxonomyweaksemilatticemulti-tasklearningtransformerfew-shotpathprediction

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper introduces TAMLEC, a method for extreme multi-label completion that models the label taxonomy as a weak semilattice and splits it into Taxonomy-Aware Tasks, each a path-closed sub-taxonomy. A transformer with shared encoder and decoder blocks plus a task-specific generator per task is trained on these parallel tasks, and at inference the known labels of a document select the relevant tasks, where missing labels are predicted by generating label paths with beam search. On three benchmark datasets the authors report that TAMLEC outperforms previous XMLCo methods on nearly all precision and NDCG metrics, with the largest margin on EURLex, and that in a simulated few-shot setting adding a new task only requires training its dedicated generator. If these results hold, the paper establishes that structuring multi-task learning around taxonomy paths extracts more from label hierarchies than tree-path or regularization-based predecessors.

What carries the argument

The central object is the Taxonomy-Aware Task (TAT): a subset of labels that itself forms a weak semilattice, is upward-closed along taxonomy paths, and together with its siblings covers the taxonomy except for the Condorcet winner root. TAMLEC predicts labels by generating paths — sequences from general to specific child labels — within each TAT relevant to a document's known labels, using a transformer with six shared encoder blocks, six shared decoder blocks, and a task-specific generator head per TAT. A width-aware label-smoothing loss scales the smoothing parameter with the number of children in the task, and beam search with score aggregation across paths and tasks produces the final label ranking.

What would settle it

Run TAMLEC on a dataset with naturally incomplete labels without adding missing ancestors during preprocessing, or test the simulated XMLCo protocol without the hierarchical-completeness step; if precision@1 on EURLex then falls below HECTOR's 0.914, the reported advantage is an artifact of the completion preprocessing rather than of the model.

Watch

Extended reading notes

Core claim

The central claim is that decomposing an extreme multi-label completion problem into Taxonomy-Aware Tasks — sub-taxonomies of a weak semilattice that preserve upward label paths and cover all non-root labels — and training a transformer with shared encoder/decoder blocks and one task-specific generator per task yields better completion than the state of the art. In the paper's experiments, TAMLEC reaches 0.945 precision@1 and 0.958 NDCG@2 on EURLex, against 0.914 and 0.926 for HECTOR, improves most metrics on MAG-CS, and statistically ties HECTOR on PubMed while remaining competitive on all baseline comparisons. TAMLEC also handles weak-semilattice taxonomies where a label has multiple parents, which tree-only path predictors such as HECTOR cannot use without removing relations.

Load-bearing premise

The method assumes that every document's label set can be completed to at least one valid taxonomy path, and silently adds missing ancestor labels during preprocessing, so if that repair changes the label distribution the reported gains may not carry over to real incomplete-label data.

Editorial extensions

If this is right

On EURLex, TAMLEC reports precision@1 of 0.945 and NDCG@2 of 0.958, versus 0.914 and 0.926 for HECTOR, a margin the paper attributes to the task decomposition.
On MAG-CS, TAMLEC improves most metrics over HECTOR (precision@1 0.870 vs 0.856), and on PubMed it matches HECTOR, indicating the gains transfer across different taxonomy shapes.
For few-shot XML, a new TAT introduced after training only needs its task-specific generator trained, and TAMLEC's global metrics stay close to full-training levels (EURLex global precision@1 0.944 vs 0.945 in the regular experiment), while baselines degrade or overfit to the new task.
Because TATs exist for weak semilattices, not just trees, the method applies to taxonomies where labels inherit from multiple parents, such as MeSH in PubMed, without the relation pruning that tree-based HECTOR requires.
The documented gap between TAMLEC and the ablated variant with reduced weight sharing and no adaptive loss shows that both the adaptive loss and the task-specific generators contribute to the reported performance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the shared encoder/decoder learns task-independent representations, swapping the GloVe text embedding for a pretrained language-model encoder could raise the ceiling further, since only the task-specific head would need retraining for a new task.
The width-aware smoothing suggests a natural curriculum — train on narrow, easy tasks first and wide, hard tasks later — which the paper does not test but follows directly from its own difficulty measure.
Deployed systems with genuinely incomplete labels would need TAMLEC's own predictions to repair the training data, creating a bootstrapping loop the paper's simulated protocol does not evaluate.
Since the TAT decomposition is unique for a given weak semilattice, an open question is how the number and width of tasks trade off against parameter sharing and generalization — the paper reports dataset-dependent outcomes but offers no formal characterization.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

TAMLEC is a genuine advance over HECTOR, and the empirical results on their chosen protocol are real. The architecture is not a patch job: weak-semilattice path prediction, taxonomy-aware tasks, and a width-dependent loss hang together coherently, and the ablation suggests each piece matters. I would send this to a serious referee without hesitation. But the main claim needs to be read with a specific caveat: the XMLCo protocol is carefully constructed so that every test document keeps only the most general labels. That means the known labels are always a valid root prefix, exactly what TAMLEC's task selection and path extension need. Real incomplete labels are messier – a document can have a specific label while its ancestors are missing, or connected non-root labels. The paper is transparent about the completeness assumption and the preprocessing, but it never tests the method under that messier regime, so the "outperforms SOTA" claim is only established for the hierarchical-completion protocol, not for XMLCo in general.

The comparison with HECTOR is mostly fair, but forcing a DAG into a tree to run HECTOR does handicap it on datasets like PubMed and EURLex. The tie on PubMed is telling: the weak-semilattice freedom alone isn't the whole story; the big EURLex margin may come from TAMLEC's many-TAT structure. The few-shot experiment strongly favors TAMLEC by design, since only its new task generator is trained, but that's also a legitimate strength of the architecture, not a trick.

The paper needs code and error bars before its numbers can be taken as settled. Single-run deep learning results without variance are hard to trust, especially on margins like 0.945 vs 0.914. The loss description also doesn't match the standard label-smoothing formula, and the uniqueness of TATs is asserted without proof. These are fixable. The stress test about the evaluation protocol is the most substantive concern, but it doesn't kill the paper because the authors are explicit about their assumptions. I would ask them to add a supplementary experiment where the observed labels are less clean (e.g., random subsets, or removing specific labels without adding ancestors) and report how TAMLEC and HECTOR behave. That would settle whether the advantage transfers.

Referee Report

5 major / 5 minor

Summary. The paper introduces TAMLEC (Taxonomy-Aware Multi-task Learning for Extreme multi-label Completion), a transformer-based method for XMLCo that decomposes a label taxonomy into Taxonomy-Aware Tasks (TATs), i.e., sub-weak-semilattices induced by the label hierarchy, and trains the model with parallel feature sharing between tasks. At inference, known labels select relevant TATs, and missing labels are predicted by extending root-to-label paths using beam search. The paper claims that TAMLEC outperforms state-of-the-art XMLCo methods on MAG-CS, PubMed, and EURLex, and is particularly effective in few-shot XML settings when new TATs appear after training. The main technical contributions are the weak-semilattice formulation of taxonomies, the TAT decomposition, the task-aware loss, and the modular transformer architecture with task-specific generators.

Significance. If the empirical claims hold, TAMLEC would be a meaningful advance for XMLCo: it handles non-tree taxonomies more faithfully than tree-based predecessors such as HECTOR, uses multi-task learning to mitigate label scarcity, and offers a natural mechanism for few-shot adaptation by only training a new task-specific generator. The evaluation is broad in terms of datasets and baselines, and the idea of decomposing the taxonomy into overlapping upper sets (TATs) is potentially useful beyond the specific architecture. However, the significance is tempered by the experimental protocol, which simulates incompleteness in a way that seems tailored to TAMLEC's path-prefix mechanism, and by the absence of error bars and the lack of a released implementation. The loss function and the uniqueness/computation of TATs also need clarification before the method is fully reproducible.

major comments (5)

[Section 4.2, 'Label Completion'] The XMLCo evaluation keeps only the most general labels per document, so every test document has an ancestor-closed set of known labels that are exactly the roots of the TATs. This is precisely the regime in which TAMLEC's task-selection and path-extension mechanism is well-defined. The paper does not evaluate on other realistic incompleteness patterns, such as missing intermediate labels or documents containing specific labels without their ancestors. Therefore the central claim that TAMLEC 'outperforms state-of-the-art methods for various XMLCo problems' is established only under this favorable protocol. Please add at least one alternative incompleteness setting, or explicitly narrow the claim.
[Section 3.2, Loss Function] The displayed loss, L = (1-α) log P(ℓ̂=ℓ) + Σ_{ℓ'≠ℓ} α log(1-P(ℓ̂=ℓ')), is not the stated label-smoothed cross-entropy. Standard label smoothing uses a term of the form ε/(K-1) Σ_{ℓ'≠ℓ} log P(ℓ̂=ℓ') (or its negative), not log(1-P(ℓ̂=ℓ')), and the formula as written has no overall minus sign, so minimizing L would maximize a sum of log-probabilities. This makes the actual training objective ambiguous. Please correct the formula or clarify how the implemented loss differs from standard label smoothing.
[Section 4.1-4.3, Tables 2 and 3] No standard deviations, confidence intervals, or significance tests are reported for any result. Several key differences are numerically small (e.g., MAG-CS P@1: 0.870 for TAMLEC vs 0.856 for HECTOR; PubMed P@1 tie at 0.866) and could fall within run-to-run variance for transformer models. The claims of consistent superiority need to be supported by multiple seeds with error margins, or at least a sensitivity analysis of the main results.
[Section 3.1, Definition 3.6] The paper states that 'the combination of the conditions in Definition 3.6 makes TATs decomposition unique' but provides no proof. It also gives no algorithm for constructing the TATs from an arbitrary weak-semilattice. Since the architecture, the loss, and the few-shot procedure all assume a fixed decomposition, the method is not fully specified without a constructive procedure or a proof of uniqueness. Please provide a proof or an explicit algorithm, and describe how the TATs were obtained for the three datasets in the experiments.
[Section 4.2, Comparison with HECTOR] The taxonomies given to HECTOR are modified by removing the minimum number of relations to reduce each weak-semilattice to a tree, while TAMLEC uses the original taxonomy. This changes the input representation for the strongest baseline and could confound the reported gains: it is unclear how much of the improvement over HECTOR comes from handling the DAG structure rather than from TAMLEC's multi-task components. Please provide an ablation, e.g., run TAMLEC on the tree-reduced taxonomies as well, or otherwise quantify the effect of the tree conversion on HECTOR's performance.

minor comments (5)

[Section 4.1] The sentence 'We will openly publish the code of our model as well as our full evaluation setup once this paper is accepted' is not sufficient for reproducibility. Please provide code or a detailed implementation description at review time.
[Section 4.1] There is a typo in the training details: 'with an initial learning rate of 5 × 10−5, a and a weight decay of 10−2' should read 'and a weight decay'.
[Section 4.2, Ablation] The ablation labeled '√TAMLEC' is described as 'a tampered down version of TAMLEC without the adaptive loss or the advanced weight sharing.' Please specify precisely which components are removed (e.g., whether the task-specific generators are retained) and whether the same hyperparameters are used.
[Section 3.2, Preprocessing] The preprocessing step that adds missing labels to form a path states that ties between equal-length paths are 'broken at random.' This introduces nondeterminism that can affect both training and inference; please specify the random seed or describe the impact of this choice.
[Section 4.3] The statement that TAMLEC's NT metrics are 'always close to the best performing method' is vague because in several cells (e.g., MAG-CS NT P@1, PubMed NT NDCG@2) the gap is substantial. Please quantify 'close' or present a trade-off criterion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: TAMLEC's empirical claims rest on independent benchmarks and disclosed design choices, not on self-referential derivation.

full rationale

The paper's central claims are empirical: TAMLEC outperforms state-of-the-art methods on XMLCo and shows strong few-shot performance. These claims are supported by experiments on three external datasets (MAG-CS, PubMed, EURLex) against multiple baselines, including an overlapping-author baseline HECTOR. No result is derived from a fitted constant that is then renamed as a prediction; the loss hyperparameters (epsilon = 0.01, learning rate = 5e-5) are fixed and the ablation model sqrt(TAMLEC) is used to test component contributions. The few-shot protocol fine-tunes all methods on the new task before evaluation, and TAMLEC's generator-only fine-tuning is disclosed rather than hidden reuse of test labels. The completeness assumption and the preprocessing that adds missing ancestors are stated explicitly; the XMLCo simulation that keeps only the most general labels is an evaluation design, not a mathematical equivalence. Even if that protocol favors TAMLEC, that is a generalization or fairness concern, not circularity. The only explicitly missing support is the deferred release of code, which affects reproducibility but does not make any derivation circular. Overall, no load-bearing step reduces to its own input by construction.

Assumptions & free parameters 3 free parameters · 3 assumptions · 1 invented entities

The central empirical claim rests on several assumptions: that the taxonomy can be modeled as a weak-semilattice, that documents can be made hierarchically complete without harming the task, and that a unique TATs decomposition exists. The method also depends on hand-set hyperparameters such as epsilon, learning rate, and weight decay, which are not fitted to data.

free parameters (3)

label smoothing epsilon = 0.01
Hand-set hyperparameter in the TAMLEC loss (Section 3.2). It is not fitted to data but controls the relative weight of the smoothing term and affects the loss scale.
learning rate = 5e-5
Adam initial learning rate for TAMLEC, stated in Section 4.1. Standard training hyperparameter, not fitted to the target metrics.
weight decay = 1e-2
Adam weight decay for TAMLEC, stated in Section 4.1. Standard regularization hyperparameter.

assumptions (3)

domain assumption The label taxonomy is a Weak-Semilattice: every subset of labels has a common lower bound.
Section 3.1, Definition 3.2. Required for the path prediction and TAT construction. Some real taxonomies may have disconnected components or no global root.
domain assumption For every document, each label's ancestor labels are present or can be added without altering the task (hierarchical completeness).
Section 3.2, training paragraph: 'missing labels are added during the preprocessing of the dataset'. This changes the data distribution and underlies the path transformation.
ad hoc to paper The TATs decomposition described by Definition 3.6 is unique and can be computed.
Section 3.1, text after Definition 3.6: 'the combination of the conditions... makes TATs decomposition unique.' No proof is given, and the definition lacks the maximality condition needed for uniqueness.

invented entities (1)

Taxonomy-Aware Tasks (TATs)
purpose: Subsets of labels used to define multi-task training and task-specific generators.
Algorithmic construct introduced in Section 3.1; no empirical handle outside the paper's own experiments.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Extreme Multi-label Completion for Semantic Document Labelling with Taxonomy-Aware Parallel Learning." pith.science (2026). https://pith.science/paper/PP7NIQPS

@misc{pith2026241213809,
  author       = {Pith},
  title        = {Pith review of: Extreme Multi-label Completion for Semantic Document Labelling with Taxonomy-Aware Parallel Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PP7NIQPS}},
  note         = {Machine review of arXiv:2412.13809}
}

read the original abstract

In Extreme Multi Label Completion (XMLCo), the objective is to predict the missing labels of a collection of documents. Together with XML Classification, XMLCo is arguably one of the most challenging document classification tasks, as the very high number of labels (at least ten of thousands) is generally very large compared to the number of available labelled documents in the training dataset. Such a task is often accompanied by a taxonomy that encodes the labels organic relationships, and many methods have been proposed to leverage this hierarchy to improve the results of XMLCo algorithms. In this paper, we propose a new approach to this problem, TAMLEC (Taxonomy-Aware Multi-task Learning for Extreme multi-label Completion). TAMLEC divides the problem into several Taxonomy-Aware Tasks, i.e. subsets of labels adapted to the hierarchical paths of the taxonomy, and trains on these tasks using a dynamic Parallel Feature sharing approach, where some parts of the model are shared between tasks while others are task-specific. Then, at inference time, TAMLEC uses the labels available in a document to infer the appropriate tasks and to predict missing labels. To achieve this result, TAMLEC uses a modified transformer architecture that predicts ordered sequences of labels on a Weak-Semilattice structure that is naturally induced by the tasks. This approach yields multiple advantages. First, our experiments on real-world datasets show that TAMLEC outperforms state-of-the-art methods for various XMLCo problems. Second, TAMLEC is by construction particularly suited for few-shots XML tasks, where new tasks or labels are introduced with only few examples, and extensive evaluations highlight its strong performance compared to existing methods.

Figures

Figures reproduced from arXiv: 2412.13809 by the authors.

**Figure 2.** Example of a Weak Semilattice taxonomy with a [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** TAMLEC’s architecture. The model is made of 6 En [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 40 canonical work pages

[1]

Gustavo Aguilar, Suraj Maharjan, Adrián Pastor López Mo nroy, and Thamar Solorio. 2017. A Multi-task Approach for Named Entity Recog nition in Social Media Data. In Proceedings of the 3rd Workshop on Noisy User-generated Text . 148–153

work page 2017
[2]

Isabelle Augenstein and Anders Søgaard. 2017. Multi-ta sk learning of keyphrase boundary classiﬁcation. arXiv preprint arXiv:1704.00514 (2017)

work page Pith review arXiv 2017
[3]

Kush Bhatia, Himanshu Jain, Purushottam Kar, Manik Varm a, and Prateek Jain

work page
[4]

Silva, Bruno Martins, and Migue l Won

Danielle Caled, Mário J. Silva, Bruno Martins, and Migue l Won

work page
[5]

Rich Caruana. 1997. Multitask learning. Machine learning 28 (1997), 41–75

1997
[6]

Ivan Chajda, Radomír Halaš, and Jan Kühr. 2007. Semilattice structures. Vol. 30. Heldermann Lemgo

work page 2007
[7]

Ilias Chalkidis, Manos Fergadiotis, and Ion Androutsop oulos. 2021. Mul- tiEURLEX – A multi-lingual and multi-label legal document c lassiﬁca- tion dataset for zero-shot cross-lingual transfer. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Process ing (Punta Cana, Dominican Republic). Association for Computational Linguist...

arXiv 2021
[8]

Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakas iotis, and Ion Androut- sopoulos. 2019. Large-scale multi-label text classiﬁcati on on EU legislation. arXiv preprint arXiv:1906.02192 (2019)

arXiv 2019

Show all 50 references

[9]

Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakas iotis, and Ion Androut- sopoulos. 2019. Large-Scale Multi-Label Text Classiﬁcati on on EU Legislation. In Proceedings of the 57th Conference of the Association for Com putational Lin- guistics, ACL 2019, Florence, Italy, Ju...

2019 doi
[10]

Shuaichen Chang, Pengfei Liu, Yun Tang, Jing Huang, Xia odong He, and Bowen Zhou. 2020. Zero-shot text-to-sql learning with auxiliary task. In Proceedings of the AAAI Conference on Artiﬁcial Intelligence , Vol. 34. 7488–7495

2020
[11]

Wei-Cheng Chang, Hsiang-Fu Yu, Kai Zhong, Yiming Yang, and Inderjit S. Dhillon. 2020. Taming Pretrained Transformers for Extreme Multi-label Text Classiﬁcation. In KDD ’20: The 26th ACM SIGKDD Conference on Knowledge Dis- covery and Data Mining, Virtual Event, CA, USA, August ...

2020
[12]

Olivier Chapelle, Pannagadatta Shivaswamy, Srinivas Vadrevu, Kilian Wein- berger, Ya Zhang, and Belle Tseng. 2010. Multi-task learnin g for boosting with application to web search ranking. In Proceedings of the 16th ACM SIGKDD in- ternational conference on Knowledge discovery ...

2010
[13]

Shijie Chen, Yu Zhang, and Qiang Yang. 2024. Multi-task learning in natural language processing: An overview. Comput. Surveys 56, 12 (2024), 1–32

2024
[14]

Xiao Chu, Wanli Ouyang, Wei Yang, and Xiaogang Wang. 201 5. Multi-task re- current neural network for immediacy prediction. In Proceedings of the IEEE international conference on computer vision . 3352–3360
[15]

Giuseppe Cuccu, Christophe Broillet, Carolin Reischa uer, Harriet Thoeny, and Philippe Cudré-Mauroux. 2022. Typhon: Parallel Transfer o n Heterogeneous Datasets for Cancer Detection in Computer-Aided Diagnosis . In 2022 IEEE In- ternational Conference on Big Data (Big Data) . ...

2022
[16]

Giuseppe Cuccu, Johan Jobin, Julien Clément, Akansha B hardwaj, Carolin Reis- chauer, Harriet Thöny, and Philippe Cudré-Mauroux. 2020. H ydra: Cancer de- tection leveraging multiple heads and heterogeneous datas ets. In 2020 IEEE In- ternational Conference on Big Data (Big Dat...

2020
[17]

Yang Deng, Yuexiang Xie, Yaliang Li, Min Yang, Nan Du, We i Fan, Kai Lei, and Ying Shen. 2019. Multi-task learning with multi-view attention for answer selec- tion and knowledge base question answering. In Proceedings of the AAAI Con- ference on Artiﬁcial Intelligence, Vol. 33...

2019
[18]

Yuxiao Dong, Hao Ma, Zhihong Shen, and Kuansan Wang. 201 7. A century of science: Globalization of scientiﬁc collaborations, ci tations, and innovations. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining . 1437–1446
[19]

Chelsea Finn, Pieter Abbeel, and Sergey Levine. 2017. M odel-agnostic meta- learning for fast adaptation of deep networks. In International conference on ma- chine learning. PMLR, 1126–1135

2017
[20]

Francesco Gargiulo, Stefano Silvestri, and Mario Ciam pi. 2019. Exploit hierar- chical label knowledge for deep learning. In 2019 IEEE 32nd International Sym- posium on Computer-Based Medical Systems (CBMS) . IEEE, 539–542

2019
[21]

Francesco Gargiulo, Stefano Silvestri, Mario Ciampi, and Giuseppe De Pietro. 2019. Deep neural network for hierarchical extreme multi- label text classiﬁcation. Appl. Soft Comput. 79 (2019), 125–138. https://doi.org/10.1016/j.asoc.2019.03.041

2019 doi
[22]

Vivek Gupta, Rahul Wadbude, Nagarajan Natarajan, Hari sh Karnick, Prateek Jain, and Piyush Rai. 2019. Distributional Semantics Meets Multi-Label Learn- ing. In The Thirty-Third AAAI Conference on Artiﬁcial Intelligenc e, AAAI 2019, The Thirty-First Innovative Applications of A...

2019
[23]

Himanshu Jain, Yashoteja Prabhu, and Manik Varma. 2016 . Extreme Multi-label Loss Functions for Recommendation, Tagging, Ranking & Othe r Missing Label Applications. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Fran...

2016
[24]

Himanshu Jain, Yashoteja Prabhu, and Manik Varma. 2016 . Extreme multi-label loss functions for recommendation, tagging, ranking & othe r missing label ap- plications. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining . 935–944

2016
[25]

Yova Kementchedjhieva and Ilias Chalkidis. 2023. An Ex ploration of Encoder- Decoder Approaches to Multi-Label Classiﬁcation for Legal and Biomedical Text. arXiv preprint arXiv:2305.05627 (2023)

2023 arXiv
[26]

Jingzhou Liu, Wei-Cheng Chang, Yuexin Wu, and Yiming Ya ng. 2017. Deep Learning for Extreme Multi-label Text Classiﬁcation. In Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval, Shinjuku, Tokyo, Japan, August 7-11, ...

2017
[27]

Jingzhou Liu, Wei-Cheng Chang, Yuexin Wu, and Yiming Ya ng. [n. d.]. Deep Learning for Extreme Multi-label Text Classiﬁcation. In Proceedings of the 40th International ACM SIGIR Conference on Research and Dev elopment in Information Retrieval (Shinjuku Tokyo Japan, 2017-08-07)...

2017
[28]

Weiwei Liu, Xiaobo Shen, Haobo Wang, and Ivor W. Tsang. 2 020. The Emerging Trends of Multi-Label Learning. CoRR abs/2011.11197 (2020)

2020 arXiv
[29]

Anshul Mittal, Kunal Dahiya, Sheshansh Agrawal, Deepa k Saini, Sumeet Agar- wal, Purushottam Kar, and Manik Varma. 2021. Decaf: Deep ext reme classiﬁca- tion with label features. In Proceedings of the 14th ACM international conference on web search and data mining . 49–57

2021
[30]

Natalia Ostapuk, Julien Audiﬀren, Ljiljana Dolamic, A lain Mermoud, and Philippe Cudré-Mauroux. 2024. Follow the Path: Hierarchy- Aware Extreme Multi-Label Completion for Semantic Text Tagging. In Proceedings of the ACM on Web Conference 2024. 2094–2105

2024
[31]

Jeﬀrey Pennington, Richard Socher, and Christopher D. Manning
[32]

Yashoteja Prabhu, Anil Kag, Shilpa Gopinath, Kunal Dah iya, Shrutendra Har- sola, Rahul Agrawal, and Manik Varma. 2018. Extreme Multi-l abel Learning with Label Features for Warm-start Tagging, Ranking & Recom mendation. In Proceedings of the Eleventh ACM International Confere...

2018
[33]

Yashoteja Prabhu and Manik Varma. 2014. FastXML: a fast , accurate and stable tree-classiﬁer for extreme multi-label learning. In The 20th ACM SIGKDD In- ternational Conference on Knowledge Discovery and Data Min ing, KDD ’14, New York, NY, USA - August 24 - 27, 2014 . ACM, 263–272

2014
[34]

Miguel Romero, Felipe Kenji Nakano, Jorge Finke, Camil o Rocha, and Celine Vens. 2023. Leveraging class hierarchy for detecting missi ng annotations on hierarchical multi-label classiﬁcation. Comput. Biol. Medicine 152 (2023), 106423. https://doi.org/10.1016/j.compbiomed.2022.106423

2023
[35]

Jake Snell, Kevin Swersky, and Richard Zemel. 2017. Pro totypical networks for few-shot learning. Advances in neural information processing systems 30 (2017). Julien Audiﬀren, Christophe Broillet, Ljiljana Dolamic, a nd Philippe Cudré-Mauroux

2017
[36]

Yisheng Song, Ting Wang, Puyu Cai, Subrota K Mondal, and Jyoti Prakash Sa- hoo. 2023. A comprehensive survey of few-shot learning: Evo lution, applica- tions, challenges, and opportunities. Comput. Surveys 55, 13s (2023), 1–40

2023
[37]

Christian Szegedy, Vincent Vanhoucke, Sergey Ioﬀe, Jo nathon Shlens, and Zbig- niew Wojna. 2016. Rethinking the Inception Architecture fo r Computer Vi- sion. In 2016 IEEE Conference on Computer Vision and Pattern Recognit ion, CVPR 2016, Las Vegas, NV, USA, June 27-30, 2016 ....

2016 doi
[38]

Yukihiro Tagami. 2017. AnnexML: Approximate Nearest N eighbor Search for Extreme Multi-label Classiﬁcation. In Proceedings of the 23rd ACM SIGKDD In- ternational Conference on Knowledge Discovery and Data Min ing, Halifax, NS, Canada, August 13 - 17, 2017 . ACM, 455–464

2017
[39]

Yik Yang Tan, Chee-Onn Chow, Jeevan Kanesan, Joon Huang Chuah, and YongLiang Lim. 2023. Sentiment analysis and sarcasm detect ion using deep multi-task learning. Wireless personal communications 129, 3 (2023), 2213–2237

2023
[40]

Gomez, Lukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko reit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. A ttention is All you Need. In Advances in Neural Information Processing Systems 30: Annual Con- ference on Neural Information Processing System...

2017
[41]

Kuansan Wang, Zhihong Shen, Chiyuan Huang, Chieh-Han W u, Yuxiao Dong, and Anshul Kanakia. 2020. Microsoft Academic Graph: When ex perts are not enough. Quant. Sci. Stud. 1, 1 (2020), 396–413

2020
[42]

Fuzhao Xue, Yao Fu, Wangchunshu Zhou, Zangwei Zheng, an d Yang You. [n. d.]. To Repeat or Not To Repeat: Insights from Scaling LLM under To ken-Crisis. https://doi.org/10.48550/arXiv.2305.13230 arXiv:2305.13230 [cs]

doi arXiv
[43]

Ronghui You, Zihan Zhang, Ziye Wang, Suyang Dai, Hirosh i Mamitsuka, and Shanfeng Zhu. 2019. AttentionXML: Label Tree-based Attent ion-Aware Deep Model for High-Performance Extreme Multi-Label Text Class iﬁcation. In Ad- vances in Neural Information Processing Systems 32: Annu...

2019
[44]

Jiong Zhang, Wei-Cheng Chang, Hsiang-Fu Yu, and Inderj it Dhillon. 2021. Fast multi-resolution transformer ﬁne-tuning for extreme mult i-label text classiﬁca- tion. Advances in Neural Information Processing Systems 34 (2021), 7267–7280

2021
[45]

Yu Zhang, Zhihong Shen, Yuxiao Dong, Kuansan Wang, and J iawei Han. 2021. MATCH: Metadata-Aware Text Classiﬁcation in A Large Hierar chy. In WWW ’21: The Web Conference 2021, Virtual Event / Ljubljana, Slovenia, April 19-23, 2021. ACM / IW3C2, 3246–3257

2021
[46]

Yu Zhang and Qiang Yang. 2018. An overview of multi-task learning. National Science Review 5, 1 (2018), 30–43

2018
[47]

Yu Zhang and Qiang Yang. 2021. A survey on multi-task lea rning. IEEE trans- actions on knowledge and data engineering 34, 12 (2021), 5586–5609

2021
[2014]

In Empir- ical Methods in Natural Language Processing (EMNLP)

GloVe: Global Vectors for Word Representation. In Empir- ical Methods in Natural Language Processing (EMNLP) . 1532–1543. http://www.aclweb.org/anthology/D14-1162
[2015]

In Ad- vances in Neural Information Processing Systems 28: Annual C onference on Neu- ral Information Processing Systems 2015, December 7-12, 201 5, Montreal, Quebec, Canada

Sparse Local Embeddings for Extreme Multi-label Clas siﬁcation. In Ad- vances in Neural Information Processing Systems 28: Annual C onference on Neu- ral Information Processing Systems 2015, December 7-12, 201 5, Montreal, Quebec, Canada. 730–738

2015
[2022]

Multi-label classiﬁcation of legislative contents w ith hierarchi- cal label attention networks. Int. J. Digit. Libr. 23, 1 (2022), 77–90. https://doi.org/10.1007/s00799-021-00307-w

2022 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Gustavo Aguilar, Suraj Maharjan, Adrián Pastor López Mo nroy, and Thamar Solorio. 2017. A Multi-task Approach for Named Entity Recog nition in Social Media Data. In Proceedings of the 3rd Workshop on Noisy User-generated Text . 148–153

work page 2017

[2] [2]

Isabelle Augenstein and Anders Søgaard. 2017. Multi-ta sk learning of keyphrase boundary classiﬁcation. arXiv preprint arXiv:1704.00514 (2017)

work page Pith review arXiv 2017

[3] [3]

Kush Bhatia, Himanshu Jain, Purushottam Kar, Manik Varm a, and Prateek Jain

work page

[4] [4]

Silva, Bruno Martins, and Migue l Won

Danielle Caled, Mário J. Silva, Bruno Martins, and Migue l Won

work page

[5] [5]

Rich Caruana. 1997. Multitask learning. Machine learning 28 (1997), 41–75

1997

[6] [6]

Ivan Chajda, Radomír Halaš, and Jan Kühr. 2007. Semilattice structures. Vol. 30. Heldermann Lemgo

work page 2007

[7] [7]

Ilias Chalkidis, Manos Fergadiotis, and Ion Androutsop oulos. 2021. Mul- tiEURLEX – A multi-lingual and multi-label legal document c lassiﬁca- tion dataset for zero-shot cross-lingual transfer. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Process ing (Punta Cana, Dominican Republic). Association for Computational Linguist...

arXiv 2021

[8] [8]

Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakas iotis, and Ion Androut- sopoulos. 2019. Large-scale multi-label text classiﬁcati on on EU legislation. arXiv preprint arXiv:1906.02192 (2019)

arXiv 2019