Pith. sign in

REVIEW 4 major objections 5 minor 81 references

Enhanced Urdu Intent Detection with Large Language Models and Prototype-Informed Predictive Pipelines

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Few-shot Urdu intent detection can label classes a model never saw in training, with a reported 98.25% F1 in the 5-shot ATIS setting.

desk verdict First few-shot Urdu intent detection benchmark, but the re-training step leaks test and validation text, so the headline F1 numbers do not measure unseen-class generalization. read the letter →

arxiv 2505.07857 v1 pith:RIG5LFAE submitted 2025-05-08 cs.CL cs.AI

classification cs.CLcs.AI
keywords Urduintentdetectionfew-shotlearningcontrastiveprototype-informedattentionunseenclasseslow-resourcelanguageprocessingmultilingualmodelsnaturalunderstanding
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

Urdu is the world's tenth most-spoken language, but prior Urdu intent detectors could only assign intents they saw during training. This paper claims to introduce LLMPIA, the first few-shot Urdu intent-detection pipeline that can also predict unseen intent classes: it re-trains a pre-trained language model on unlabeled Urdu text with a contrastive objective, then a prototype-informed attention network classifies a query by comparing it to a few labeled examples per intent. On the Urdu ATIS and Web Queries benchmarks the pipeline reports 83.28% and 98.25% F1 (4-way 1-shot and 5-shot) on ATIS, and 76.23% and 84.42% F1 on Web Queries, with a same-class case study reporting a 53.55% F1 gain over the strongest existing Urdu predictor. If the result holds, developers of Urdu conversational systems would have a practical route to adding new intents from one to five examples per class instead of large labeled corpora.

What carries the argument

The load-bearing mechanism has two stages. First, LLMRCL (Large Language Models Enhanced Representations with Contrastive Learning) re-trains a pre-trained encoder on unlabeled Urdu text with two objectives: masked language modelling at a 25% mask rate, plus a self-supervised contrastive loss that pulls semantically related sentence pairs together and pushes unrelated pairs apart. Second, Prototype-Informed Attention (PIA) builds on a Siamese architecture, meaning two network branches that share the same weights, to turn the encoded support examples into a class prototype and to compare a query embedding against that prototype. The attention layers compute inner-sentence and inner-class feature interactions, a prototype-informed layer refines the prototype, and an adaptive layer maps prototypes and queries into a common space before cosine similarity with a temperature coefficient scores the match. Two unsupervised contrastive regularization terms are added to the cross-entropy metric loss, so the total objective simultaneously polishes prototypes and prevents few-shot overfitting.

What would settle it

A direct falsifier is to rerun the full LLMPIA evaluation after replacing the combined-split unlabeled corpus with unlabeled text drawn only from the training split; if the 4-way 1-shot and 4-way 5-shot F1 scores on ATIS and Web Queries drop materially, the original unseen-class numbers depend on exposure to test text, not on generalization. A weaker check is to search the re-training corpus for test queries and measure how much of the test set appears in it verbatim.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a language model's Urdu representations can be sharpened for few-shot intent detection without any labeled data, by re-training it with a self-supervised contrastive loss on unlabeled Urdu text, and that those sharper embeddings plug directly into a prototype-informed attention mechanism that compares queries with class prototypes built from a small support set. The paper reports that this combined pipeline, LLMPIA, predicts unseen intent classes from seen classes: 83.28% F1 in 4-way 1-shot and 98.25% F1 in 4-way 5-shot on the Urdu ATIS dataset, and 76.23% and 84.42% F1 on the Urdu Web Queries dataset. Across all six pre-trained language models and both datasets, the contrastively re-trained versions outperform their pre-trained counterparts. In the unseen-class few-shot regime the best encoder is the South-Asian multilingual MuRIL model, and cosine similarity is the most reliable of the thirteen similarity measures tested. In a Web Queries case study where all classes appear in both training and test sets, the paper reports that its pipeline exceeds the strongest existing Urdu predictor by 53.55% F1.

Load-bearing premise

Section 4.3 of the paper states that during LLMCRL re-training the authors combined the train, test, and validation sets of each dataset; the few-shot unseen-class results therefore rest on the premise that using test and validation text in this unlabeled re-training does not leak information about the unseen classes.

Editorial extensions

If this is right

  • In the few-shot unseen-class regime, a multilingual model trained on South Asian languages outperforms the Urdu-only model, so the best encoder choice depends on whether the deployment must handle new intents.
  • Re-training always helps: every one of the six encoders improves after contrastive re-training on unlabeled Urdu text, making the re-training stage a reliable component rather than a lucky exception.
  • The 13-metric study shows cosine similarity is the safe default for prototype-based Urdu intent matching, while Hamming distance and KL divergence collapse to near-zero F1 in the same settings.
  • Adding support examples helps: 4-way 5-shot consistently beats 4-way 1-shot, so few-shot Urdu intent detection still rewards collecting a handful of labeled examples per intent.
  • If the reported unseen-class performance is genuine, Urdu moves from having no few-shot intent detector to having a method whose reported F1 scores are competitive with those for higher-resource languages.

Reading between the lines

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

  • A testable extension the paper does not run: repeat LLMRCL using only training-split unlabeled text and compare the F1 gap; if it shrinks, some of the reported unseen-class gain comes from exposure to test and validation text rather than from generalization.
  • The recipe is not Urdu-specific: any low-resource language with unlabeled text but few intent labels could use the same contrastive re-training plus prototype-attention steps, though the paper does not test this.
  • For deployed assistants, the 1-shot numbers imply that one user-confirmed utterance could open a new intent class, making the support set a live product operation rather than a fixed training artifact.
  • Because the strongest few-shot encoder is multilingual rather than Urdu-only, the same prototype pipeline may benefit from simply swapping in newer multilingual encoders as they appear, without changing the rest of the architecture.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes LLMPIA, a few-shot Urdu intent detection pipeline that combines contrastive re-training of pre-trained language models (LLMCRL) with a prototype-informed attention mechanism (PIA). It evaluates six pre-trained language models and thirteen similarity measures on two Urdu datasets, ATIS and Urdu Web Queries, under 4-way 1-shot and 4-way 5-shot settings, and also compares against existing Urdu intent detection predictors. The abstract claims headline F1 scores of 83.28% and 98.25% on ATIS and 76.23% and 84.42% on Web Queries, plus a 53.55% F1-Score improvement over the state of the art.

Significance. If the evaluation were valid, the systematic comparison of six PLMs and thirteen similarity measures on two Urdu datasets would be a useful resource for low-resource Urdu NLP. The manuscript is transparent about hyperparameter search and early stopping, and it covers a broad configuration space. However, as submitted, the central claims are not supported: the LLMCRL re-training stage uses the test and validation text, so the reported few-shot results do not measure inductive generalization to unseen classes. In addition, the abstract misreports which experimental settings produced the headline numbers, and the claimed 53.55% improvement over the state of the art is not present in the results. These issues require a substantial rework of the experimental evaluation and the reported claims.

major comments (4)
  1. [Section 4.3, Eq. (1), Section 3] Section 4.3 states: 'During the re-training of models in the LLMCRL, we combined train, test and validation sets of each dataset.' Since Eq. (1) defines L_LLMCRL = L_MLM(D_unlabeled; θ) + L_SCL(D_unlabeled; θ), and Section 3 defines D_unlabeled as the text from all categories, the encoder used in Tables 5–8 is optimized on the same test and validation utterances that later form the support and query sets described in Section 3.2. The reported F1 scores therefore reflect transductive adaptation to the evaluation split rather than inductive prediction of unseen classes from seen classes. Because the paper's central claim is about generalization to unseen classes, this leakage invalidates the headline few-shot results.
  2. [Abstract vs. Tables 5 and 6] The abstract reports '83.28% and 76.23% F1-Score' under the 4-way 1-shot setting for ATIS and Web Queries, respectively. These values correspond to the MuRIL pre-trained 4-way 5-shot results at 75% seen classes in Table 5 (F1 = 0.8328) and Table 6 (F1 = 0.7623), not to any 1-shot result. The actual MuRIL re-trained 1-shot F1 scores at the 75% split are 0.9581 (ATIS) and 0.8403 (Web Queries). The abstract's pairing of experimental settings with numbers is therefore incorrect and must be corrected.
  3. [Abstract vs. Table 9, Section 7] The abstract claims that LLMPIA 'outperformed state-of-the-art predictor by 53.55% F1-Score' on the Web Queries dataset under same-class train/test settings. This figure does not appear in Table 9 or anywhere in the text. Table 9 shows the best proposed pipeline (roberta-small-Urdu, 3-way 5-shot) achieving F1 = 0.9555 versus U-IntentCapsNet's 0.9084, a relative improvement of about 4.71%, which the text itself states. The 53.55% claim should be either substantiated with a precise definition and calculation or removed.
  4. [Section 6, Tables 3–8] Because the validation set is included in LLMCRL re-training and is also used for early stopping and hyperparameter selection, the model selection reported in Tables 3–4 and the similarity-metric rankings in Tables 7–8 are affected by the same leakage. Furthermore, the paper reports no variance or confidence intervals across episodes or random splits, so it is not possible to assess whether the differences between models or similarity measures that drive the 'best pipeline' conclusions are statistically meaningful.
minor comments (5)
  1. [Section 4.1] The abbreviation for the contrastive re-training stage is inconsistent: Section 4.1 uses 'LLMRCL' while Sections 4.3 and elsewhere use 'LLMCRL'; please standardize throughout.
  2. [Header] The line 'Preprint submitted to Nuclear Physics B' appears to be a template artifact and should be removed for a computer-science journal submission.
  3. [Tables 7 and 8] The MuRIL model is labeled 'MURiL-base-100-languages' in Table 7 and 'Muril-base-100-languages' in Table 8, while Table 2 and the text correctly call it 'MuRIL-base-17-languages'; the labels should be made consistent.
  4. [Section 6.1] The text quotes accuracy values for MuRIL (e.g., 0.2375, 0.2731) that do not appear in Table 6; please reconcile the prose with the reported tables.
  5. [Figure 2] Figure 2 shows identical four-intent examples under both 'Web-Queries Dataset' and 'ATIS Dataset' labels, which is confusing; the figure should clarify which examples come from which dataset.

Circularity Check

1 steps flagged · score 6.0 of 10

LLMCRL re-trains the encoders on the test and validation splits, so the reported few-shot F1 scores do not measure unseen-class prediction from seen classes alone.

  1. fitted input called prediction [Section 3 (D_unlabeled definition), Section 4.3 'Training Process', and Tables 5-8]
    ""During the re-training of models in the LLMCRL, we combined train, test and validation sets of each dataset." ... "We similarly extract text from the validation set Dval and test set Dtest to construct support and query sets." ... "L_LLMCRL = L_MLM(D_unlabeled; θ) + L_SCL(D_unlabeled; θ) (1)""

    By the paper's own definitions, D_unlabeled (the LLMCRL retraining corpus) includes Dval and Dtest, since Call contains Cunlabeled,unknown and Dval/Dtest are drawn from that category; Section 4.3 confirms all three splits are combined. Equation (1) optimizes θ on these texts through masked language modeling and self-supervised contrastive learning. Section 3.2 then constructs the few-shot validation/testing support and query sets from those same Dval/Dtest texts, and Tables 5-8 report F1 on them. The encoder is therefore representation-fitted to the exact evaluation utterances before the unseen-class prediction is made, so the headline scores are not predictions from seen classes alone.

full rationale

The central claimed novelty is few-shot prediction of unseen intent classes from seen classes. That evaluation claim reduces, at the critical step, to fitting the representation learner on the test/validation text and then measuring prediction on that same text: Section 4.3 explicitly combines train, test and validation sets for LLMCRL, and Sections 3 and 3.2 define the evaluation support/query sets from Dval/Dtest while Eq. (1) trains on D_unlabeled. This is a fitted-input-called-prediction pattern and is load-bearing because the abstract's headline F1 numbers come from this protocol. I do not find additional circularity: the PIA module is adapted from Snell et al. (an external prototypical-network reference), the similarity-metric comparison is an empirical sweep rather than a derived result, and there is no self-citation chain used to force the architecture. The score is 6 rather than higher because class labels are still unseen and the episode training does use seen-class examples, so some genuine transfer is being measured; nonetheless, the reported generalization to 'unseen' classes is partially circular by construction.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central numbers depend on grid-selected hyperparameters, an undisclosed temperature, and the premise that unlabeled test data can be used in re-training. No new physical or ontological entities are introduced; the PIA attention module is presented as borrowed from [61], though the citation is mismatched.

free parameters (7)
  • LLMCRL learning rate = 1e-5 or 1e-6 per model/dataset
    Selected by grid search over {1e-3,...,2e-6} in Tables 3/4; affects all downstream F1 scores.
  • LLMCRL epochs = 10 to 200 depending on model
    Early stopping on validation makes the effective epoch count a fitted quantity.
  • PIA learning rate = 1e-4 to 1e-6 per model
    Grid-searched in Tables 3/4.
  • PIA dropout rate = 0.1
    Optimal value from search space {0.05,...,0.5}.
  • PIA hidden size = 300
    Optimal value from {100,...,500}.
  • Temperature coefficients tau and t = Not reported
    Equations 8 and 9 include temperature parameters that strongly affect contrastive loss and similarity scaling; no values are disclosed in the paper.
  • MLM masking rate = 25%
    Chosen in Section 4.1; 80/10/10 mask-random-unchanged split is a design choice.
assumptions (5)
  • domain assumption Pre-trained multilingual and Urdu-specific language models provide transferable representations for Urdu intent detection.
    The entire framework assumes PLM embeddings are a useful starting point before re-training (Section 4).
  • domain assumption Translating English ATIS into Urdu with Google Translate plus human review yields a valid Urdu intent detection benchmark.
    Section 5.1 relies on this translated dataset without reporting translation quality metrics or human agreement.
  • ad hoc to paper Using unlabeled text from test and validation splits during re-training does not invalidate few-shot generalization claims.
    Section 4.3 explicitly combines train, test and validation sets for LLMCRL; this premise is load-bearing and contested.
  • domain assumption Cosine similarity is an appropriate metric for comparing prototypes and query embeddings.
    Used in Eq. 9 and recommended after testing 13 metrics; the choice is empirical but the default assumption is that cosine captures intent semantics.
  • domain assumption Self-supervised contrastive learning with shuffled, stop-word-removed sentences preserves intent-relevant semantics.
    Section 4.1 describes the augmentation; no validation that shuffling does not destroy intent cues.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhanced Urdu Intent Detection with Large Language Models and Prototype-Informed Predictive Pipelines." pith.science (2026). https://pith.science/paper/RIG5LFAE

@misc{pith2026250507857,
  author       = {Pith},
  title        = {Pith review of: Enhanced Urdu Intent Detection with Large Language Models and Prototype-Informed Predictive Pipelines},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RIG5LFAE}},
  note         = {Machine review of arXiv:2505.07857}
}
read the original abstract

Multifarious intent detection predictors are developed for different languages, including English, Chinese and French, however, the field remains underdeveloped for Urdu, the 10th most spoken language. In the realm of well-known languages, intent detection predictors utilize the strategy of few-shot learning and prediction of unseen classes based on the model training on seen classes. However, Urdu language lacks few-shot strategy based intent detection predictors and traditional predictors are focused on prediction of the same classes which models have seen in the train set. To empower Urdu language specific intent detection, this introduces a unique contrastive learning approach that leverages unlabeled Urdu data to re-train pre-trained language models. This re-training empowers LLMs representation learning for the downstream intent detection task. Finally, it reaps the combined potential of pre-trained LLMs and the prototype-informed attention mechanism to create a comprehensive end-to-end LLMPIA intent detection pipeline. Under the paradigm of proposed predictive pipeline, it explores the potential of 6 distinct language models and 13 distinct similarity computation methods. The proposed framework is evaluated on 2 public benchmark datasets, namely ATIS encompassing 5836 samples and Web Queries having 8519 samples. Across ATIS dataset under 4-way 1 shot and 4-way 5 shot experimental settings LLMPIA achieved 83.28% and 98.25% F1-Score and on Web Queries dataset produced 76.23% and 84.42% F1-Score, respectively. In an additional case study on the Web Queries dataset under same classes train and test set settings, LLMPIA outperformed state-of-the-art predictor by 53.55% F1-Score.

Figures

Figures reproduced from arXiv: 2505.07857 by the authors.

Figure 1
Figure 1. Graphical illustration of Key Steps of Large Language Models Enhanced Repre [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Graphical illustration of the key steps in the Prototype Informed Attention [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Graphical overview of dataset statistics for Urdu intent detection [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: A Comprehensive Performance Analysis of Distinct Predictive Pipelines of Pro [PITH_FULL_IMAGE:figures/full_fig_p025_4.png]
Figure 5
Figure 5. Figure 5: A Comprehensive Performance Analysis of Distinct Predictive Pipelines of Pro [PITH_FULL_IMAGE:figures/full_fig_p026_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

81 extracted references · 68 canonical work pages

  1. [1]

    Con- trastive multi-graph learning with neighbor hierarchical sifting for semi- supervised text classification

    Ai, W., Li, J., Wang, Z., Wei, Y., Meng, T., Li, K., 2025. Con- trastive multi-graph learning with neighbor hierarchical sifting for semi- supervised text classification. Expert Systems with Applications 266, 125952

  2. [2]

    Survey on self-supervised learning: Auxiliary pretext tasks and contrastive learning methods in imaging

    Albelwi, S., 2022. Survey on self-supervised learning: Auxiliary pretext tasks and contrastive learning methods in imaging. Entropy 24, 551

  3. [3]

    Access structures for angular similarity queries

    Apaydin, T., Ferhatosmanoglu, H., 2006. Access structures for angular similarity queries. IEEE transactions on knowledge and data engineering 18, 1512–1525

  4. [4]

    Benchmarking performance of machine and deep learning-based methodologies for urdu text document classifica- tion

    Asim, M.N., Ghani, M.U., Ibrahim, M.A., Mahmood, W., Dengel, A., Ahmed, S., 2021. Benchmarking performance of machine and deep learning-based methodologies for urdu text document classifica- tion. Neural Computing and Applications 33, 5437–5469

  5. [5]

    Adh-ppi: An attention-based deep hybrid model for protein-protein in- teraction prediction

    Asim, M.N., Ibrahim, M.A., Malik, M.I., Dengel, A., Ahmed, S., 2022a. Adh-ppi: An attention-based deep hybrid model for protein-protein in- teraction prediction. iScience 25

  6. [6]

    El-rmlocnet: An explainable lstm network for rna-associated multi-compartment localization prediction

    Asim, M.N., Ibrahim, M.A., Malik, M.I., Zehe, C., Cloarec, O., Trygg, J., Dengel, A., Ahmed, S., 2022b. El-rmlocnet: An explainable lstm network for rna-associated multi-compartment localization prediction. Computational and Structural Biotechnology Journal 20, 3986–4002

  7. [7]

    Effect of pruning on feature ranking metrics in highly skewed datasets in text classification

    Asim, M.N., Rehman, A., Idrees, M., 2017. Effect of pruning on feature ranking metrics in highly skewed datasets in text classification. Interna- tional Journal of Computer Science and Network Security 17, 135–144

  8. [8]

    Zero-shot retrieval of user intent in human-robot interaction with large language models

    Atuhurra, J., 2024. Zero-shot retrieval of user intent in human-robot interaction with large language models. IEEE International Conference on Multimedia Information Processing and Retrieval (MIPR) . 33

Show all 81 references
  1. [9]

    Implementation of a virtual assistant system based on deep multi-modal data integration

    Baek, S., Kim, J., Lee, J., Lee, M., 2024. Implementation of a virtual assistant system based on deep multi-modal data integration. Journal of Signal Processing Systems 96, 179–189

  2. [10]

    Neural multi-task learning for end-to-end arabic aspect-based sentiment analysis

    Bensoltane, R., Zaki, T., 2025. Neural multi-task learning for end-to-end arabic aspect-based sentiment analysis. Computer Speech & Language 89, 101683

  3. [11]

    On-device llms for home assistant: Dual role in intent detection and response generation

    Birkmose, R., Reece, N.M., Norvin, E.H., Bjerva, J., Zhang, M., 2025. On-device llms for home assistant: Dual role in intent detection and response generation. arXiv preprint arXiv:2502.12923

  4. [12]

    Generalized hamming distance

    Bookstein, A., Kulyukin, V.A., Raita, T., 2002. Generalized hamming distance. Information Retrieval 5, 353–375

  5. [13]

    Efficient intent detection with dual sentence encoders

    Casanueva, I., Temˇ cinas, T., Gerz, D., Henderson, M., Vuli´ c, I., 2020. Efficient intent detection with dual sentence encoders. arXiv preprint arXiv:2003.04807

  6. [14]

    An intent recogni- tion pipeline for conversational ai

    Chandrakala, C., Bhardwaj, R., Pujari, C., 2024. An intent recogni- tion pipeline for conversational ai. International Journal of Information Technology 16, 731–743

  7. [15]

    Cheng, X., Zhu, Z., Yang, B., Zhuang, X., Li, H., Zou, Y., 2024. Cyclical contrastive learning based on geodesic for zero-shot cross-lingual spoken language understanding, in: Findings of the Association for Computa- tional Linguistics: ACL 2024, pp. 1806–1816

  8. [16]

    Debiased contrastive learning

    Chuang, C.Y., Robinson, J., Lin, Y.C., Torralba, A., Jegelka, S., 2020. Debiased contrastive learning. Advances in Neural Information Process- ing Systems 33, 8765–8775

  9. [17]

    Chebyshev distance

    Coghetto, R., 2016. Chebyshev distance. Formalized Mathematics 24, 121–141

  10. [18]

    Pearson correlation coefficient

    Cohen, I., Huang, Y., Chen, J., Benesty, J., Benesty, J., Chen, J., Huang, Y., Cohen, I., 2009. Pearson correlation coefficient. Noise re- duction in speech processing , 1–4

  11. [19]

    Zero-shot- bert-adapters: A zero-shot pipeline for unknown intent detection, in: 34 Findings of the Association for Computational Linguistics: EMNLP 2023, pp

    Comi, D., Christofidellis, D., Piazza, P., Manica, M., 2023. Zero-shot- bert-adapters: A zero-shot pipeline for unknown intent detection, in: 34 Findings of the Association for Computational Linguistics: EMNLP 2023, pp. 650–663

  12. [20]

    Euclidean distance mapping

    Danielsson, P.E., 1980. Euclidean distance mapping. Computer Graph- ics and image processing 14, 227–248

  13. [21]

    Devlin, J., Chang, M., Lee, K., Toutanova, K., 2019. Bert: Pre-training of deep bidirectional transformers for language understanding, in: Pro- ceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics, Minneapolis, Minnesot...

  14. [22]

    An emotion role min- ing approach based on multiview ensemble learning in social networks

    Du, Y., Wang, Y., Hu, J., Li, X., Chen, X., 2022. An emotion role min- ing approach based on multiview ensemble learning in social networks. Information Fusion 88, 100–114

  15. [23]

    A linguistics-based ap- proach to refining automatic intent detection in conversational agent design

    Ferrera, A., Mezzotero, G., Ursino, D., 2025. A linguistics-based ap- proach to refining automatic intent detection in conversational agent design. Information Sciences 689, 121493

  16. [24]

    Grattafiori, A., Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al- Dahle, A., Letman, A., Mathur, A., Schelten, A., Vaughan, A., et al.,

  17. [25]

    Exploring large language mod- els for the generation of synthetic training samples for aspect-based sen- timent analysis in low resource settings

    Hellwig, N.C., Fehle, J., Wolff, C., 2025. Exploring large language mod- els for the generation of synthetic training samples for aspect-based sen- timent analysis in low resource settings. Expert Systems with Applica- tions 261, 125514

  18. [26]

    Ghs-net a generic hybridized shallow neural network for multi- label biomedical text classification

    Ibrahim, M.A., Khan, M.U.G., Mehmood, F., Asim, M.N., Mahmood, W., 2021. Ghs-net a generic hybridized shallow neural network for multi- label biomedical text classification. Journal of biomedical informatics 116, 103699

  19. [27]

    On the jaccard similarity test

    Ivchenko, G., Honov, S., 1998. On the jaccard similarity test. Journal of Mathematical Sciences 88, 789–794

  20. [28]

    A survey on contrastive self-supervised learning

    Jaiswal, A., Babu, A.R., Zadeh, M.Z., Banerjee, D., Makedon, F., 2020. A survey on contrastive self-supervised learning. Technologies 9, 2. 35

  21. [29]

    Intent detection for task-oriented conversational agents: A comparative study of recurrent neural networks and transformer models

    Jbene, M., Chehri, A., Saadane, R., Tigani, S., Jeon, G., 2025. Intent detection for task-oriented conversational agents: A comparative study of recurrent neural networks and transformer models. Expert Systems 42, e13712

  22. [30]

    Kathakali, M., Santosh, A.A.V., Soumya, T., Aruna, M., 2023. Effect of pivot language and segment-based few-shot prompting for cross-domain multi-intent identification in low resource languages, in: Proceedings of the 20th International Conference on Natural Language Processin...

  23. [31]

    The bhattacharyya distance and detection between markov chains

    Kazakos, D., 1978. The bhattacharyya distance and detection between markov chains. IEEE Transactions on Information Theory 24, 747–754

  24. [32]

    Muril: Multilingual representations for indian languages

    Khanuja, S., Bansal, D., Mehtani, S., Khosla, S., Dey, A., Gopalan, B., Margam, D.K., Aggarwal, P., Nagipogu, R.T., Dave, S., et al., 2021. Muril: Multilingual representations for indian languages. arXiv preprint arXiv:2103.10730

  25. [33]

    Adversarial self-supervised con- trastive learning

    Kim, M., Tack, J., Hwang, S.J., 2020. Adversarial self-supervised con- trastive learning. Advances in Neural Information Processing Systems 33, 2983–2994

  26. [34]

    Kobayashi, E., Fushimi, T., Saito, K., Ikeda, T., 2014. Similarity search by generating pivots based on manhattan distance, in: PRICAI 2014: Trends in Artificial Intelligence: 13th Pacific Rim International Confer- ence on Artificial Intelligence, Gold Coast, QLD, Australia, D...

  27. [35]

    Con- versational agents in healthcare: A systematic review

    Laranjo, L., Dunn, A.G., Tong, H.L., Kocaballi, A.B., Chen, J., Bashir, R., Surian, D., Gallego, B., Magrabi, F., Lau, A.Y., et al., 2018. Con- versational agents in healthcare: A systematic review. Journal of the American Medical Informatics Association 25, 1248–1258

  28. [36]

    Conversational agents

    Lester, J., Branting, K., Mott, B., 2004. Conversational agents. The Practical Handbook of Internet Computing , 220–240

  29. [37]

    Logistic regression matching pursuit algorithm for text classification

    Li, Q., Zhao, S., Zhao, S., Wen, J., 2023. Logistic regression matching pursuit algorithm for text classification. Knowledge-Based Systems 277, 110761. 36

  30. [38]

    Ud bbc: Named entity recognition in social network combined bert-bilstm-crf with active learning

    Li, W., Du, Y., Li, X., Chen, X., Xie, C., Li, H., Li, X., 2022. Ud bbc: Named entity recognition in social network combined bert-bilstm-crf with active learning. Engineering Applications of Artificial Intelligence 116, 105460

  31. [39]

    Named entity recognition for smart city data streams: Enhancing visualization and interaction

    Li, X., Wang, J., Zhang, L., 2025. Named entity recognition for smart city data streams: Enhancing visualization and interaction. Authorea Preprints

  32. [40]

    Roberta: A robustly optimized bert pretraining approach

    Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., Stoyanov, V., 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692

  33. [41]

    Lu, H.y., Zhang, J.m., Du, Y., Xia, C., Wang, C., Fang, W., Wu, X.j.,

  34. [42]

    Low-resource lan- guages: A review of past work and future challenges

    Magueresse, A., Carles, V., Heetderks, E., 2020. Low-resource lan- guages: A review of past work and future challenges. arXiv preprint arXiv:2006.07264

  35. [43]

    A parameter-free text classification method based on dual compressors

    Mao, Y., Ding, Y., Cui, T., 2025. A parameter-free text classification method based on dual compressors. Knowledge and Information Systems , 1–31

  36. [44]

    Passion-net: A robust precise and explainable predic- tor for hate speech detection in roman urdu text

    Mehmood, F., Ghafoor, H., Asim, M.N., Ghani, M.U., Mahmood, W., Dengel, A., 2024. Passion-net: A robust precise and explainable predic- tor for hate speech detection in roman urdu text. Neural Computing and Applications 36, 3077–3100

  37. [45]

    A precisely xtreme-multi channel hybrid ap- proach for roman urdu sentiment analysis

    Mehmood, F., Ghani, M.U., Ibrahim, M.A., Shahzadi, R., Mahmood, W., Asim, M.N., 2020. A precisely xtreme-multi channel hybrid ap- proach for roman urdu sentiment analysis. IEEE Access 8, 192740– 192759

  38. [46]

    Enml: multi-label ensemble learn- ing for urdu text classification

    Mehmood, F., Shahzadi, R., Ghafoor, H., Asim, M.N., Ghani, M.U., Mahmood, W., Dengel, A., 2023. Enml: multi-label ensemble learn- ing for urdu text classification. ACM Transactions on Asian and Low- Resource Language Information Processing 22, 1–31. 37

  39. [47]

    Recent advances in natural language processing via large pre-trained language models: A survey

    Min, B., Ross, H., Sulem, E., Veyseh, A.P.B., Nguyen, T.H., Sainz, O., Agirre, E., Heintz, I., Roth, D., 2023. Recent advances in natural language processing via large pre-trained language models: A survey. ACM Computing Surveys 56, 1–40

  40. [48]

    Aspect-based sentiment analysis: A dual-task learning architecture using imbalanced maximized-area under the curve proximate support vector machine and reinforcement learning

    Motevalli, M.M., Sohrabi, M.K., Yaghmaee, F., 2025. Aspect-based sentiment analysis: A dual-task learning architecture using imbalanced maximized-area under the curve proximate support vector machine and reinforcement learning. Information Sciences 689, 121449

  41. [49]

    Muennighoff, N., Wang, T., Sutawika, L., Roberts, A., Biderman, S., Scao, T.L., Bari, M.S., Shen, S., Yong, Z.X., Schoelkopf, H., et al.,

  42. [50]

    Dsl-driven approaches and metamodels for chat- bot development: A systematic literature review

    Ouaddi, C., Benaddi, L., Bouziane, E.M., Jakimi, A., Chehri, A., Saadane, R., 2025. Dsl-driven approaches and metamodels for chat- bot development: A systematic literature review. Expert Systems 42, e13787

  43. [51]

    Divide-aggregate heterogeneous hypergraph for large-scale user intention detection

    Qu, M., Song, X., Di, D., Su, T., 2024. Divide-aggregate heterogeneous hypergraph for large-scale user intention detection. Knowledge-Based Systems , 111994

  44. [52]

    Language models are unsupervised multitask learners

    Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., Sutskever, I., et al., 2019. Language models are unsupervised multitask learners. Ope- nAI blog 1, 9

  45. [53]

    Semantic co- sine similarity, in: Proceedings of the 7th International Student Con- ference on Advanced Science and Technology (ICAST), University of Seoul, Seoul, South Korea

    Rahutomo, F., Kitasuka, T., Aritsugi, M., et al., 2012. Semantic co- sine similarity, in: Proceedings of the 7th International Student Con- ference on Advanced Science and Technology (ICAST), University of Seoul, Seoul, South Korea. p. 1

  46. [54]

    Benchmark- ing with a language model initial selection for text classification tasks

    Riyadi, A., Kovacs, M., Serd¨ ult, U., Kryssanov, V., 2025. Benchmark- ing with a language model initial selection for text classification tasks. Machine Learning and Knowledge Extraction 7, 3

  47. [55]

    A characterization of random variables with minimum l2-distance

    R¨ uschendorf, L., Rachev, S.T., 1990. A characterization of random variables with minimum l2-distance. Journal of multivariate analysis 32, 48–54. 38

  48. [56]

    Fnreq-net: A hybrid computational framework for functional and non-functional requirements classification

    Saleem, S., Asim, M.N., Van Elst, L., Dengel, A., 2023. Fnreq-net: A hybrid computational framework for functional and non-functional requirements classification. Journal of King Saud University-Computer and Information Sciences 35, 101665

  49. [57]

    Mlr- predictor: a versatile and efficient computational framework for multi- label requirements classification

    Saleem, S., Asim, M.N., Van Elst, L., Junker, M., Dengel, A., 2024. Mlr- predictor: a versatile and efficient computational framework for multi- label requirements classification. Frontiers in Artificial Intelligence 7, 1481581

  50. [58]

    Improving user intent detection in urdu web queries with capsule net architectures

    Shams, S., Aslam, M., 2022. Improving user intent detection in urdu web queries with capsule net architectures. Applied Sciences 12, 11861

  51. [59]

    Shams, S., Aslam, M., Martinez-Enriquez, A.M., 2019. Lexical intent recognition in urdu queries using deep neural networks, in: Advances in soft computing: 18th Mexican International Conference on Artificial Intelligence, MICAI 2019, Xalapa, Mexico, October 27–November 2, 2019...

  52. [60]

    Regularization with dot- product kernels

    Smola, A., Ov´ ari, Z., Williamson, R.C., 2000. Regularization with dot- product kernels. Advances in neural information processing systems 13

  53. [61]

    Prototypical networks for few- shot learning, in: Proceedings of the 31st International Conference on Neural Information Processing Systems, Red Hook, NY, USA

    Snell, J., Swersky, K., Zemel, R., 2017. Prototypical networks for few- shot learning, in: Proceedings of the 31st International Conference on Neural Information Processing Systems, Red Hook, NY, USA. pp. 4080– 4090

  54. [62]

    Circnet: An encoder-decoder-based convolution neural network (cnn) for circular rna identification

    Stricker, M., Asim, M.N., Dengel, A., Ahmed, S., 2022. Circnet: An encoder-decoder-based convolution neural network (cnn) for circular rna identification. Neural Computing and Applications , 1–12

  55. [63]

    Summra, S., Usman, M.G., Muhammad, A., et al., 2021. Supervised neural network for offline forgery detection of handwritten signature, in: 2021 18th International Conference on Electrical Engineering, Comput- ing Science and Automatic Control (CCE), IEEE. pp. 1–6

  56. [64]

    Bench- marking pre-trained large language models’ potential across urdu nlp tasks

    Tahir, M.H., Shams, S., Fiaz, L., Adeeba, F., Hussain, S., 2024. Bench- marking pre-trained large language models’ potential across urdu nlp tasks. arXiv preprint arXiv:2405.15453 . 39

  57. [65]

    Un ministral, des ministraux

    Team, M.A., 2024. Un ministral, des ministraux. Mistral AI News URL: https://mistral.ai/news/ministraux/. accessed: 2024-10-29

  58. [66]

    Llama 2: Open foundation and fine-tuned chat models

    Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., Bhosale, S., et al., 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288

  59. [67]

    R´ enyi divergence and kullback- leibler divergence

    Van Erven, T., Harremos, P., 2014. R´ enyi divergence and kullback- leibler divergence. IEEE Transactions on Information Theory 60, 3797– 3820

  60. [68]

    Lexical paraphrasing and pseudo relevance feedback for biomedical document retrieval

    Wasim, M., Asim, M.N., Ghani, M.U., Rehman, Z.U., Rho, S., Mehmood, I., 2019. Lexical paraphrasing and pseudo relevance feedback for biomedical document retrieval. Multimedia Tools and Applications 78, 29681–29712

  61. [69]

    Winkler, M., Juozapaityte, V., Van Der Goot, R., Plank, B., 2024. Slot and intent detection resources for bavarian and lithuanian: Assessing translations vs natural queries to digital assistants, in: Proceedings of the 2024 Joint International Conference on Computational Lingu...

  62. [70]

    Are all languages created equal in multilingual bert? arXiv preprint arXiv:2005.09093

    Wu, S., Dredze, M., 2020. Are all languages created equal in multilingual bert? arXiv preprint arXiv:2005.09093

  63. [71]

    Xu, Z., Cruz, M.J., Guevara, M., Wang, T., Deshpande, M., Wang, X., Li, Z., 2024. Retrieval-augmented generation with knowledge graphs for customer service question answering, in: Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Informa...

  64. [72]

    Intent-enhanced attentive bert capsule network for zero-shot intention detection

    Xue, S., Ren, F., 2021. Intent-enhanced attentive bert capsule network for zero-shot intention detection. Neurocomputing 458, 1–13

  65. [73]

    Ltner: Large language model tagging for named entity recognition with contextualized entity marking, in: International Conference on Pattern Recognition, Springer

    Yan, F., Yu, P., Chen, X., 2025. Ltner: Large language model tagging for named entity recognition with contextualized entity marking, in: International Conference on Pattern Recognition, Springer. pp. 399–411. 40

  66. [74]

    Gfidf: gradual fusion intent detection framework

    Yang, Q., Liu, Y., Lu, L., Liu, L., 2025. Gfidf: gradual fusion intent detection framework. The Journal of Supercomputing 81, 312

  67. [75]

    A comprehensive capability analysis of gpt-3 and gpt-3.5 series models

    Ye, J., Chen, X., Xu, N., Zu, C., Shao, Z., Liu, S., Cui, Y., Zhou, Z., Gong, C., Shen, Y., et al., 2023. A comprehensive capability analysis of gpt-3 and gpt-3.5 series models. arXiv preprint arXiv:2303.10420

  68. [76]

    Dice similarity measure between single valued neutrosophic multisets and its application in medical diagnosis

    Ye, S., Ye, J., 2014. Dice similarity measure between single valued neutrosophic multisets and its application in medical diagnosis. Neutro- sophic sets and systems 6, 9

  69. [77]

    Midlm: Multi-intent detection with bidirectional large language models, in: Proceedings of the 31st Inter- national Conference on Computational Linguistics, pp

    Yin, S., Huang, P., Xu, Y., 2025. Midlm: Multi-intent detection with bidirectional large language models, in: Proceedings of the 31st Inter- national Conference on Computational Linguistics, pp. 2616–2625

  70. [78]

    New intent dis- covery with pretraining and contrastive learning, in: Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics, Dublin, Ireland

    Zhang, Y., Zhang, H., Zhan, L., Wu, X., Lam, A., 2022. New intent dis- covery with pretraining and contrastive learning, in: Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics, Dublin, Ireland. pp. 256–269. 41

  71. [2022]

    arXiv preprint arXiv:2211.01786

    Crosslingual generalization through multitask finetuning. arXiv preprint arXiv:2211.01786

  72. [2024]

    arXiv preprint arXiv:2407.21783

    The llama 3 herd of models. arXiv preprint arXiv:2407.21783

  73. [2025]

    Neurocomputing , 129554

    Enhancing few-shot out-of-distribution intent detection by reduc- ing attention misallocation. Neurocomputing , 129554

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.