Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

Unified Large Language Models for Misinformation Detection in Low-Resource Linguistic Settings

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a fused 27,568-item Urdu news benchmark, LUND, supports a majority-vote ensemble of six pre-trained LLMs that reaches 0.959 accuracy and 0.960 F1, beating individual models and prior ML/DL systems.

desk verdict A useful Urdu dataset merge undermined by an unverifiable label mapping and inaccessible data. read the letter →

arxiv 2506.01587 v1 pith:763EOBRZ submitted 2025-06-02 cs.CL

Muhammad Islam , Javed Ali Khan , Mohammed Abaker , Ali Daud , Azeem Irshad This is my paper
classification cs.CL
keywords Urdufakenewsdetectionlow-resourceNLPlargelanguagemodelsbenchmarkdatasetensemblestackingmisinformationmultilingualtransformerstextclassification
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

The paper sets out to close a resource gap: fake-news detection for Urdu, a language with hundreds of millions of speakers, has lacked a large public benchmark, so methods cannot be compared on common ground. It proposes LUND, a corpus of 27,568 Urdu news items created by merging three existing public datasets, standardizing their labels into a binary legit/fake split, deduplicating entries, and balancing classes across 23 domains. On this corpus, the paper evaluates six pre-trained multilingual language models and then combines their predictions by majority-vote stacking into ULLM, which it reports at 0.959 accuracy and 0.960 F1, above every individual model and above the machine-learning and deep-learning systems used previously. If true, the contribution is twofold: a standard benchmark for future Urdu fake-news research, and a strong ensemble baseline that low-resource language work can reuse.

What carries the argument

The load-bearing object is LUND, the fused corpus, and the load-bearing mechanism is ULLM, a stacking ensemble. Each of six pre-trained multilingual transformers independently produces class predictions (legit or fake) for a news item, and the ensemble takes a majority vote across the six outputs; the paper also uses tokenizers and embeddings matched to each model, fine-tunes hyperparameters (e.g., 10 epochs, batch size 16–32, learning rate $4 \times 10^{-10}$), and applies preprocessing including stemming, stop-word removal, and TF-IDF/n-gram features for classical classifiers. The dataset construction—standardizing inconsistent labels (including mapping 'True', 'Partly True', and 'Half True' into fake), removing duplicates, and balancing classes—is what makes the subsequent model comparison meaningful.

What would settle it

Re-label a random sample of, say, 500 LUND items with two independent native-Urdu annotators who are shown only the news text, then compare their labels with the dataset's and re-run the ULLM evaluation on the items where they agree; a large drop below the reported 0.959 accuracy and 0.960 F1 would show the results ride on faulty ground truth.

Watch

Extended reading notes

Core claim

The central discovery on the paper's own terms is that merging three publicly available Urdu news collections and cleaning them into one balanced, deduplicated, domain-diverse corpus yields the first large benchmark fake-news dataset for Urdu, LUND, with 14,185 fake and 13,383 legit items. The paper then shows that a unified model formed by majority-vote stacking of six pre-trained transformers—XLNet, XLM-RoBERTa, mBERT, RoBERTa, DistilBERT, and DeBERTa—classifies this dataset with 0.959 accuracy, 0.961 precision, 0.958 recall, and 0.960 F1, outperforming each base model, all compared machine-learning and deep-learning baselines, and prior Urdu fake-news systems. The authors additionally claim that expert journalists vetted samples before and after model evaluation, supporting the dataset's reliability.

Load-bearing premise

The results all rest on the assumption that the labels inherited from the three merged datasets are accurate and mutually consistent after the paper's standardization, including its mapping of 'True', 'Partly True', and 'Half True' into fake news.

Editorial extensions

If this is right

  • Urdu fake-news research now has a single 27,568-sample benchmark with fixed train/test splits (22,055 training, 5,513 testing), enabling direct comparison of future systems.
  • The proposed ULLM provides a strong ensemble baseline at 0.959 accuracy and 0.960 F1 that any new Urdu fake-news detector can be measured against.
  • Pre-trained multilingual language models, especially when combined, outperform classical machine learning and single deep-learning models on this corpus, suggesting transfer learning works for Urdu news text.
  • Because the corpus spans 23 domains, models trained on it should generalize better than the domain-specific Urdu datasets used previously.
  • The dataset can support downstream tools such as Urdu fact-checking algorithms and further evaluation with other metrics and embeddings.

Reading between the lines

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

  • An untested implication is that the ensemble's margin over its best single model (XLM-RoBERTa at 0.925 F1) could shrink or vanish if the disputed relabeling of 'Partly True' and 'Half True' as fake is revised; a sensitivity analysis over label definitions would settle this.
  • The ensemble's gains may partly reflect disagreement among multilingual models rather than Urdu-specific understanding; evaluating on Roman Urdu or on a held-out native-Urdu-only test set would separate those effects.
  • The same stacking recipe could be transferred to other low-resource languages with comparable fused datasets, but only if the source label quality is independently checked first.
  • The paper reports no per-domain accuracy, so an editorial extension would be to test whether the 23-domain corpus hides weak spots in underrepresented domains such as women's rights, religion, education, and health.
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.

Referee Report

5 major / 6 minor

Summary. The paper claims to curate LUND, a 27,568-item Urdu fake news dataset by merging three existing public datasets, and to propose ULLM, a majority-vote stack of six pre-trained LLMs, reporting 0.959 accuracy and 0.960 F1. The authors further claim that LUND is the first benchmark large fake news detection dataset for Urdu, that it is publicly available via GitHub, and that expert journalists vetted the samples. The paper also compares six LLMs and several ML/DL baselines on LUND.

Significance. If LUND were properly constructed, publicly released, and the label mapping were correct, the corpus could serve as a useful benchmark for Urdu fake news detection, a genuinely under-served area. The comparison of multiple multilingual pre-trained models is also potentially informative. However, as written, the central claims are not supported: the label standardization appears to corrupt the ground truth, the dataset is not actually publicly available despite the claim, the 'first benchmark' claim is contradicted by the paper's own literature table, and the evaluation lacks statistical grounding. The paper therefore does not currently provide a reliable benchmark or a defensible model comparison.

major comments (5)
  1. [III-B.4] The label standardization step is load-bearing and, as written, appears to corrupt the ground truth. The sentence 'FND is a binary class problem, so we transform True, Partly True, and Half True labels into fake news' maps truth-valued labels to the fake class. If taken literally, every model comparison in Tables 4 and 5 is built on systematically wrong labels. If this is a typo, the manuscript must state the correct mapping (e.g., False, Partly False, Half False) and release the mapping code. No inter-annotator agreement, human verification protocol, or label-noise analysis is provided, so the abstract's 'expert-verified' claim is unsupported.
  2. [Data Availability] The manuscript contradicts itself on data availability. The abstract and Section I state that the dataset and code are publicly available at https://github.com/MislamSatti/Urdu-Large-language-dataset, but the DATA AVAILABILITY section says 'The data source will be provided on demand.' This contradiction prevents independent verification of the dataset and undermines the claimed public benchmark status. The authors should either provide a working public repository with the dataset and preprocessing code or state clearly that the data is not public.
  3. [I and Table 1] The claim that LUND is 'the first benchmark large FND dataset for Urdu' is contradicted by the paper's own Table 1, which lists 'Ax-to-Grind Urdu: Benchmark Dataset for Urdu Fake News Detection' (TrustCom 2024) and the UrduFake@FIRE2020 shared task. These are existing Urdu fake news benchmarks. The novelty claim needs to be substantially revised, for example by positioning LUND as the first fused multi-domain large Urdu dataset, not the first benchmark.
  4. [Tables 4 and 5] The experimental evaluation reports a single 80/20 split with no error bars, no confidence intervals, and no significance tests. The proposed ULLM is a majority vote of the same six baseline models evaluated on the same test set, so the aggregation rule is not independently validated. To support the claim that ULLM 'outperforms' all baselines, the authors should provide multiple runs, cross-validation, or a statistical significance test, and explain how the aggregation rule avoids overfitting to the test set.
  5. [Abstract and III-C] The dataset statistics are internally inconsistent. The abstract states 'total words of 106,221K and 84.07K unique words,' while Section III-C states 'The overall dataset contains 67.91k words in both fake and legitimate Urdu news text.' The conclusion also gives 27,410 instances instead of the 27,568 reported elsewhere. Such numerical inconsistencies make it difficult to trust the reported corpus statistics and should be reconciled with the actual released dataset.
minor comments (6)
  1. [III-E] The text says 'We have chosen five LLM models' but then enumerates six models (XLNet, mBERT, XLM-RoBERTa, DistilBERT, RoBERTa, DeBERTa). Section III-F.1 additionally lists GPT-2 among the selected models. The final model set used in ULLM is therefore unclear.
  2. [III-E, Eq. (2)] The masked language modeling objective in Eq. (2) is written as a product over all tokens with no masking indicator; this is not the standard MLM objective. Please correct the equation or add the appropriate masking notation.
  3. [III-F.3] The learning rate is reported as 4 × 10^-10, which is implausibly small for fine-tuning transformer models. Please report the actual learning rate, schedule, and optimizer settings used.
  4. [II] The sentence 'Journalists and experts wet the sample annotated dataset' should read 'vetted' rather than 'wet'.
  5. [References] Reference [8] is cited as the LIAR dataset, but the cited paper is a psychology study on children's evaluation of lies. The LIAR dataset should be cited to Wang (2017).
  6. [III-D.1 and III-D.2] The sentence 'Figure 5 showcases the techniques used in tokenizer and encoded objects' appears at the end of both Section III-D.1 and Section III-D.2; one of these should refer to the correct figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark evaluation and ensemble results are self-contained empirical comparisons, not derivations from their own inputs.

full rationale

The paper's central claims are (1) construction of LUND, a 27,568-item Urdu news dataset by merging three existing public datasets, and (2) that a majority-vote ensemble (ULLM) of six pretrained LLMs reaches 0.959 accuracy and 0.960 F1 on a held-out 20% split. Neither claim reduces to a fitted input renamed as a prediction. The ensemble prediction is obtained by training each constituent LLM separately and aggregating their predictions by majority vote; comparing the ensemble to its own constituent models is a standard ensemble evaluation and does not make the outcome equivalent to a fitted constant. The paper fits no parameter that is later presented as a derived benchmark property. Self-citations in the reference list (e.g., [36], [37], [38]) appear in generic methodological contexts such as LLM pretraining, stacking, and hyperparameter tuning, and they are not load-bearing for the benchmark or the reported performance. The statement in Section III-B.4 that "we transform True, Partly True, and Half True labels into fake news" is a serious data-labeling concern that affects dataset validity, but it is a correctness/verifiability issue, not a circularity issue: it does not make a predicted quantity identical to an input by construction. The data-availability contradiction (abstract says public GitHub, Data Availability says "on demand") similarly affects reproducibility, not circular reasoning. Because no load-bearing step in the claimed derivation chain is self-definitional, fitted-input-called-prediction, or dependent on a self-citation chain, the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper's central benchmark claims depend on the trustworthiness of labels inherited from three public datasets, a nonstandard relabeling rule, and the transferability of pretrained multilingual models to Urdu. These are domain assumptions rather than derived facts. No invented entities are introduced. The three stated hyperparameters are hand-chosen and under-specified, so the reported performance cannot be independently reproduced from the text.

free parameters (3)
  • learning_rate = 4e-10
    Hand-set in Section III-F.3; no grid search, sensitivity analysis, or explanation for an unusually low value.
  • batch_size = 16 to 32
    Hand-set range in Section III-F.3; no search or ablation.
  • training_epochs = 10
    Hand-set in Section III-F.3; no early stopping details or epoch sensitivity.
assumptions (5)
  • domain assumption The three source datasets have accurate labels from authentic sources.
    Section III-A claims sources were verified and curated, but no verification protocol, inter-annotator agreement, or label audit is provided.
  • ad hoc to paper True, Partly True, and Half True labels can be merged into the fake class without changing meaning.
    Section III-B.4 applies this mapping; it is nonstandard and can corrupt ground truth if partial truths are treated as fake.
  • domain assumption Fine-tuning pretrained multilingual transformers on 22,055 Urdu training samples transfers well enough for classification.
    Invoked throughout Section III-E; no evidence is given for the Urdu-specific readiness of each selected model.
  • domain assumption Removing stop words and stemming improves or at least does not harm fake news classification.
    Section III-D applies this pipeline without experiments comparing cleaned versus raw text.
  • domain assumption A single 80/20 random split gives a reliable performance estimate.
    Section III-C and Table III define the split; no repeated splits, cross-validation, or confidence intervals are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unified Large Language Models for Misinformation Detection in Low-Resource Linguistic Settings." pith.science (2026). https://pith.science/paper/763EOBRZ

@misc{pith2026250601587,
  author       = {Pith},
  title        = {Pith review of: Unified Large Language Models for Misinformation Detection in Low-Resource Linguistic Settings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/763EOBRZ}},
  note         = {Machine review of arXiv:2506.01587}
}
read the original abstract

The rapid expansion of social media platforms has significantly increased the dissemination of forged content and misinformation, making the detection of fake news a critical area of research. Although fact-checking efforts predominantly focus on English-language news, there is a noticeable gap in resources and strategies to detect news in regional languages, such as Urdu. Advanced Fake News Detection (FND) techniques rely heavily on large, accurately labeled datasets. However, FND in under-resourced languages like Urdu faces substantial challenges due to the scarcity of extensive corpora and the lack of validated lexical resources. Current Urdu fake news datasets are often domain-specific and inaccessible to the public. They also lack human verification, relying mainly on unverified English-to-Urdu translations, which compromises their reliability in practical applications. This study highlights the necessity of developing reliable, expert-verified, and domain-independent Urdu-enhanced FND datasets to improve fake news detection in Urdu and other resource-constrained languages. This paper presents the first benchmark large FND dataset for Urdu news, which is publicly available for validation and deep analysis. We also evaluate this dataset using multiple state-of-the-art pre-trained large language models (LLMs), such as XLNet, mBERT, XLM-RoBERTa, RoBERTa, DistilBERT, and DeBERTa. Additionally, we propose a unified LLM model that outperforms the others with different embedding and feature extraction techniques. The performance of these models is compared based on accuracy, F1 score, precision, recall, and human judgment for vetting the sample results of news.

Figures

Figures reproduced from arXiv: 2506.01587 by the authors.

Figure 1
Figure 1. The key Steps Performed in Methodology. A. Contributions This research study endeavors to assess the reliability and generalization factor of Large Urdu news dataset, in general. Moreover, it attempts to determine the capability of Large language model to perform in detecting fake news for low resource language. Specifically, the following are the major contributions of the research. • We tackle key challenges in fa… view at source ↗
Figure 2
Figure 2. word cloud of (a) legit (b) fake and Urdu news. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Our overall Proposed framework of ULLM on the LUND dataset [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Sample data of fake and legit Urdu News We also removed the duplicate entries during the merging of the dataset to avoid skewed results. Slightly altered news was also present in the merged dataset, so deduplication was hard, but we removed that news as well. C. Datase…
Figure 5
Figure 5. Figure 5: Stemming and stop word removal technique [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Demonstration of tokenizer and encoded objects [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Demonstrates our proposed unified model. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Demonstrates the comparison of all the LLMs with our [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: llustrates the comparison of all ML and DL model. [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DeBERTa-Sentinel: Toward Transparent and Trustworthy Detection of AI-Generated Text

    cs.CL 2026-08 conditional novelty 4.0 of 10

    A fine-tuned DeBERTa-v3 classifier reportedly beats a RoBERTa baseline at detecting paraphrased AI text (97.53% accuracy, 99.53% AUC), but sample-count mismatches and a possible train/test content overlap make the exa...

Reference graph

Works this paper leans on

38 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [5]

    Ax-to-Grind Urdu: Benchmark Dataset for Urdu Fake News Detection

    S. Harris, J. Liu, H. J. Hadi, and Y . Cao, “Ax-to-Grind Urdu: Benchmark Dataset for Urdu Fake News Detection”

    work page

  2. [1]

    Fake news detection on Pakistani news using machine learning and deep learning,

    A. Kishwar and A. Zafar, “Fake news detection on Pakistani news using machine learning and deep learning,” Expert Syst. Appl., vol. 211, p. 118558, Jan. 2023, doi: 10.1016/j.eswa.2022.118558

    work page arXiv 2023

  3. [2]

    Urdu- Fake@FIRE2020: Shared Track on Fake News Identification in Urdu,

    M. Amjad, G. Sidorov, A. Zhila, A. Gelbukh, and P. Rosso, “Urdu- Fake@FIRE2020: Shared Track on Fake News Identification in Urdu,” in Forum for Information Retrieval Evaluation, Hyderabad India: ACM, Dec. 2020, pp. 37–40. doi: 10.1145/3441501.3441541

    arXiv 2020

  4. [3]

    Ur Rehman Ahmed, A

    N. Ur Rehman Ahmed, A. Badshah, H. Adeel, A. Tajammul, A. Daud and T. Alsahfi, ”Visual Deepfake Detection: Review of Techniques, Tools, Limitations, and Future Prospects,” in IEEE Access, vol. 13, pp. 1923-1961, 2025, doi: 10.1109/ACCESS.2024.3523288

    arXiv 1923

  5. [4]

    RumorLLM: A Rumor Large Language Model-Based Fake-News-Detection Data-Augmentation Approach,

    J. Lai et al., “RumorLLM: A Rumor Large Language Model-Based Fake-News-Detection Data-Augmentation Approach,” Appl. Sci., vol. 14, no. 8, p. 3532, Apr. 2024, doi: 10.3390/app14083532

    doi:10.3390/app14083532 2024

  6. [6]

    & Che, D

    Daud, A., Khan, W. & Che, D. Urdu language processing: a survey. Artif Intell Rev 47, 279–311 (2017). https://doi.org/10.1007/s10462-016- 9482-x

    work page doi:10.1007/s10462-016- 2017

  7. [7]

    Discerning truth from deception: Human judgments and automation efforts,

    V . L. Rubin and N. Conroy, “Discerning truth from deception: Human judgments and automation efforts,” First Monday, Mar. 2012, doi: 10.5210/fm.v17i3.3933

    work page doi:10.5210/fm.v17i3.3933 2012

  8. [8]

    The roles of liar inten- tion, lie content, and theory of mind in children’s evaluation of lies,

    H. Cheung, T.-S. C. Siu, and L. Chen, “The roles of liar inten- tion, lie content, and theory of mind in children’s evaluation of lies,” J. Exp. Child Psychol., vol. 132, pp. 1–13, Apr. 2015, doi: 10.1016/j.jecp.2014.12.002

    work page doi:10.1016/j.jecp.2014.12.002 2015

Show all 38 references
  1. [9]

    Fake News or Truth? Using Satirical Cues to Detect Potentially Misleading News,

    V . Rubin, N. Conroy, Y . Chen, and S. Cornwell, “Fake News or Truth? Using Satirical Cues to Detect Potentially Misleading News,” in Proceedings of the Second Workshop on Computational Approaches to Deception Detection, San Diego, California: Association for Compu- tational L...

    2016 doi

  2. [10]

    Fake News Detection Through Multi-Perspective Speaker Profiles

    Y . Long, Q. Lu, R. Xiang, M. Li, and C.-R. Huang, “Fake News Detection Through Multi-Perspective Speaker Profiles”

  3. [11]

    Fake news detection in Urdu language using machine learning,

    M. S. Farooq, A. Naseem, F. Rustam, and I. Ashraf, “Fake news detection in Urdu language using machine learning,” PeerJ Comput. Sci., vol. 9, p. e1353, May 2023, doi: 10.7717/peerj-cs.1353

    2023 doi

  4. [12]

    Khan, W., Daud, A., Khan, K., Muhammad, S., & Haq, R. (2023). Exploring the frontiers of deep learning and natural language processing: A comprehensive overview of key challenges and emerging trends. Natural Language Processing Journal, 4, 100026

    2023

  5. [13]

    & Daud, A

    Kazi, S., Khoja, S. & Daud, A. A survey of deep learning techniques for machine reading comprehension. Artif Intell Rev 56 (Suppl 2), 2509–2569 (2023). https://doi.org/10.1007/s10462-023-10583-4

    2023 doi

  6. [14]

    Deciphering Deception: Unmask- ing Fake News in Multilingual Contexts,

    A. Agarwal, Y . P. Singh, and V . Rai, “Deciphering Deception: Unmask- ing Fake News in Multilingual Contexts,” in 2024 IEEE International Conference on Computing, Power and Communication Technologies (IC2PCT), Greater Noida, India: IEEE, Feb. 2024, pp. 807–812. doi: 10.1109/I...

    2024

  7. [15]

    Palani, B., & Elango, S. (2023). BBC-FND: An ensemble of deep learning framework for textual fake news detection. Computers and Electrical Engineering, 110, 108866

    2023

  8. [16]

    Raja, E., Soni, B., & Borgohain, S. K. (2024). Harnessing heterogeneity: A multi-embedding ensemble approach for detecting fake news in Dra- vidian languages. Computers and Electrical Engineering, 120, 109661

    2024

  9. [17]

    Fake News Identifica- tion in Urdu Tweets Using Machine Learning Models,

    Z. Iqbal, F. M. Khan, I. U. Khan, and I. U. Khan, “Fake News Identifica- tion in Urdu Tweets Using Machine Learning Models,” Asian Bull. Big Data Manag., vol. 4, no. 1, Feb. 2024, doi: 10.62019/abbdm.v4i1.105

    2024 doi

  10. [18]

    Enriching Urdu NER with BERT Embedding, Data Augmentation, and Hybrid Encoder- CNN Architecture,

    A. Ahmed, D. Huang, S. Y . Arafat, and I. Hameed, “Enriching Urdu NER with BERT Embedding, Data Augmentation, and Hybrid Encoder- CNN Architecture,” ACM Trans. Asian Low-Resour. Lang. Inf. Process., vol. 23, no. 4, pp. 1–38, Apr. 2024, doi: 10.1145/3648362

    2024 doi

  11. [19]

    Benchmark Performance of Machine And Deep Learning Based Methodologies for Urdu Text Document Classification

    M. N. Asim, M. U. Ghani, M. A. Ibrahim, S. Ahmad, and A. Den- gel, “Benchmark Performance of Machine And Deep Learning Based Methodologies for Urdu Text Document Classification”

  12. [20]

    Knowledge discovery through directed probabilistic topic models: a survey

  13. [21]

    Urdu Fake News Detection Using Ensemble of Machine Learning Models

    A. Hegde and H. L. Shashirekha, “Urdu Fake News Detection Using Ensemble of Machine Learning Models”

  14. [22]

    Arabic Fake News Detection in Social Media Context Using Word Embeddings and Pre-trained Transformers,

    M. Azzeh, A. Qusef, and O. Alabboushi, “Arabic Fake News Detection in Social Media Context Using Word Embeddings and Pre-trained Transformers,” Arab. J. Sci. Eng., Apr. 2024, doi: 10.1007/s13369-024- 08959-x

    2024 doi

  15. [23]

    B., & Ibor, A

    Okunoye, O. B., & Ibor, A. E. (2022). Hybrid fake news detection tech- nique with genetic search and deep learning. Computers and Electrical Engineering, 103, 108344

    2022

  16. [24]

    BiL-FaND: leveraging ensemble tech- nique for efficient bilingual fake news detection,

    S. Munir and M. Asif Naeem, “BiL-FaND: leveraging ensemble tech- nique for efficient bilingual fake news detection,” Int. J. Mach. Learn. Cybern., Mar. 2024, doi: 10.1007/s13042-024-02128-0

    2024 doi

  17. [25]

    Meeting the challenge: A benchmark corpus for automated Urdu meeting summarization,

    B. Sadia, F. Adeeba, S. Shams, and K. Javed, “Meeting the challenge: A benchmark corpus for automated Urdu meeting summarization,” Inf. Process. Manag., vol. 61, no. 4, p. 103734, Jul. 2024, doi: 10.1016/j.ipm.2024.103734

    2024

  18. [26]

    Detection of vio- lence incitation expressions in Urdu tweets using convolutional neural network,

    M. S. Khan, M. S. I. Malik, and A. Nadeem, “Detection of vio- lence incitation expressions in Urdu tweets using convolutional neural network,” Expert Syst. Appl., vol. 245, p. 123174, Jul. 2024, doi: 10.1016/j.eswa.2024.123174

    2024

  19. [27]

    Hate Speech Detection in Roman Urdu using Machine Learning Techniques,

    S. Nasir, A. Seerat, and M. Wasim, “Hate Speech Detection in Roman Urdu using Machine Learning Techniques,” in 2024 5th International Conference on Advancements in Computational Sci- ences (ICACS), Lahore, Pakistan: IEEE, Feb. 2024, pp. 1–7. doi: 10.1109/ICACS60934.2024.10473250

    2024

  20. [28]

    Massively Multilingual Corpus of Sentiment Datasets and Multi-faceted Sentiment Classification Benchmark

    Ł. Augustyniak et al., “Massively Multilingual Corpus of Sentiment Datasets and Multi-faceted Sentiment Classification Benchmark”

  21. [29]

    Detection of Sarcasm in Urdu Tweets Using Deep Learning and Transformer Based Hybrid Approaches,

    M. E. Hassan, M. Hussain, I. Maab, U. Habib, M. A. Khan, and A. Masood, “Detection of Sarcasm in Urdu Tweets Using Deep Learning and Transformer Based Hybrid Approaches,” IEEE Access, vol. 12, pp. 61542–61555, 2024, doi: 10.1109/ACCESS.2024.3393856

    2024

  22. [30]

    On the transferability of pre-trained language models for low-resource programming languages,

    F. Chen, F. H. Fard, D. Lo, and T. Bryksin, “On the transferability of pre-trained language models for low-resource programming languages,” in Proceedings of the 30th IEEE/ACM International Conference on Program Comprehension, Virtual Event: ACM, May 2022, pp. 401–412. doi: 10...

    2022

  23. [31]

    Fake news detection in low-resource languages: A novel hybrid summarization approach,

    J. Alghamdi, Y . Lin, and S. Luo, “Fake news detection in low-resource languages: A novel hybrid summarization approach,” Knowl.-Based Syst., p. 111884, May 2024, doi: 10.1016/j.knosys.2024.111884

    2024

  24. [32]

    Contextual Embeddings based on Fine-tuned Urdu-BERT for Urdu threatening content and target identification,

    M. S. I. Malik, U. Cheema, and D. I. Ignatov, “Contextual Embeddings based on Fine-tuned Urdu-BERT for Urdu threatening content and target identification,” J. King Saud Univ. - Comput. Inf. Sci., vol. 35, no. 7, p. 101606, Jul. 2023, doi: 10.1016/j.jksuci.2023.101606

    2023

  25. [33]

    Toward the Development of Large-Scale Word Embedding for Low- Resourced Language,

    S. Nazir, M. Asif, S. A. Sahi, S. Ahmad, Y . Y . Ghadi, and M. H. Aziz, “Toward the Development of Large-Scale Word Embedding for Low- Resourced Language,” IEEE Access, vol. 10, pp. 54091–54097, 2022, doi: 10.1109/ACCESS.2022.3173259

    2022

  26. [34]

    Multi-class sentiment analysis of urdu text using multilingual BERT,

    L. Khan, A. Amjad, N. Ashraf, and H.-T. Chang, “Multi-class sentiment analysis of urdu text using multilingual BERT,” Sci. Rep., vol. 12, no. 1, p. 5436, Mar. 2022, doi: 10.1038/s41598-022-09381-9

    2022 doi

  27. [35]

    UQA: Corpus for Urdu Question Answering

    S. Arif, S. Farid, A. Athar, and A. A. Raza, “UQA: Corpus for Urdu Question Answering”

  28. [36]

    Ontology-Based News Linking for Semantic Tem- poral Queries,

    M. Islam Satti et al., “Ontology-Based News Linking for Semantic Tem- poral Queries,” Comput. Mater. Contin., vol. 74, no. 2, pp. 3913–3929, 2023, doi: 10.32604/cmc.2023.033001

    2023

  29. [37]

    Depression Detection Based on Hybrid Deep Learning SSCL Framework Using Self-Attention Mechanism,

    A. Nadeem, M. Naveed, M.Satti, H. Afzal, T. Ahmad, and K.-I. Kim, “Depression Detection Based on Hybrid Deep Learning SSCL Framework Using Self-Attention Mechanism,” Sensors, vol. 22, no. 24, p. 9775, Dec. 2022, doi: 10.3390/s22249775

    2022 doi

  30. [38]

    J. A. Khan, T. Ullah, A. A. Khan, A. Yasin, M. A. Akbar, & K. Aurangzeb, ”Can end-user feedback in social media be trusted for software evolution: Exploring and analyzing fake reviews,” Concurrency and Computation: Practice and Experience, 36(10),e7990, 2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.