Pith. sign in

REVIEW 4 major objections 4 minor 31 references

Improving Answer Extraction in Context-based Question Answering Systems Using LLMs

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper claims that fine-tuning a transformer for span prediction on SQuAD1.1 turns weak zero-shot models into reliable answer extractors, with Roberta-base reaching 86.84% ROUGE-L.

desk verdict Paper's abstract and Table III say Roberta-base is best; the Conclusion says Albert-base and Bert-base are best—an internal contradiction that poisons the central claim. read the letter →

arxiv 2606.06197 v3 pith:KMKEIWOR submitted 2026-06-04 cs.CL cs.AI

Hafez Abdelghaffar , Ahmed Alansary , Ali Hamdi This is my paper · ORCID
classification cs.CLcs.AI
keywords QuestionAnsweringAnswerExtractionFine-tuningSQuAD1.1ROUGE-LBERTScoreSpanPredictionTransformerEncoders
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 claims that task-specific fine-tuning on a benchmark extractive QA dataset is what makes language models reliable at pulling answers out of a passage, and that one shared span-prediction setup can reveal which architectures are best suited. It reports that fine-tuned Roberta-base outperforms eleven other models, including larger decoder-style LLMs, achieving ROUGE-L 86.84%, BLEU 28.24%, and BERTScore 95.38%. The broader point is that pre-training strength alone does not predict QA performance; adaptability to the task matters. If true, this gives practitioners a simple recipe—concatenate context and question, add a start/end span head, fine-tune—for building accurate extractive QA systems.

What carries the argument

The load-bearing piece is the answer-span head: the model assigns probability distributions over start and end token positions, trained by the sum of two cross-entropy losses, L = L_start + L_end. At inference, the span between the highest-probability start and end indices is extracted from the context, so answers remain grounded in the passage. The paper applies this objective to transformer encoders, using the concatenated context and question as input. This span-prediction setup is what converts the models from weak zero-shot generators into extractive QA systems without needing retrieval or external knowledge.

What would settle it

Re-run all twelve models on SQuAD1.1 with a per-model hyperparameter search (learning rate, batch size, epochs) under the same data and metrics. If any model matches or exceeds Roberta-base's ROUGE-L 86.84%, BLEU 28.24%, and BERTScore 95.38% under tuned settings, the paper's identical-setup comparison is not reliable evidence that Roberta-base is the best extractive QA model.

Watch

Extended reading notes

Core claim

On SQuAD1.1, fine-tuning a transformer with a span-extraction objective (predicting start and end positions of the answer within the context) substantially improves every model tested. After fine-tuning, Roberta-base reaches the top scores on all three metrics, followed closely by Albert-base-v2 and Bert-base. The paper also finds that the ranking of models flips after fine-tuning: StableLM-2 and Qwen2.5 are the best baselines, but encoder-based masked-language models overtake them once adapted. The conclusion is that selecting a verbatim span from context is best done with a supervised span-prediction objective, and that such fine-tuning is essential for reliable and consistent outputs.

Load-bearing premise

The load-bearing premise is that one identical fine-tuning recipe is a fair test for twelve models of very different sizes and families; if the recipe happens to favor some architectures, the reported ranking could change.

Editorial extensions

If this is right

  • Fine-tuning on SQuAD1.1 dramatically lifts extractive QA performance, taking Roberta-base from 3.34% to 86.84% ROUGE-L.
  • Task-specific adaptation matters more than raw pre-training scale: smaller encoder models surpass larger general-purpose LLMs after fine-tuning.
  • Model capacity still constrains gains, as Bert-tiny lags far behind larger models even after the same fine-tuning.
  • The combination of ROUGE-L, BLEU, and BERTScore provides complementary signals: high BERTScore with lower lexical scores indicates semantically relevant but not verbatim answers.
  • The unified fine-tuning setup offers a practical template for comparing architectures on extractive QA and selecting a strong default model.

Reading between the lines

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

  • Because every SQuAD1.1 answer is a contiguous span, the same span-loss recipe should transfer directly to other extractive reading-comprehension benchmarks with minimal changes—an implication the paper leaves untested.
  • The consistent gap between high BERTScore and lower ROUGE-L/BLEU hints that answers are often semantically right but lexically off; using BERTScore to rerank candidate spans at inference could be a cheap accuracy boost the paper does not explore.
  • The paper does not report hyperparameter settings, so a per-model learning-rate, batch-size, or epoch sweep is the natural next experiment to test whether the reported leader board is stable or an artifact of the single shared recipe.
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 / 4 minor

Summary. The paper proposes a context-based QA system that fine-tunes twelve transformer models of varying sizes (4.4M to 1.7B parameters) on SQuAD1.1, formulated as extractive span prediction with a start/end cross-entropy loss. The authors evaluate baselines and fine-tuned models using ROUGE-L, BLEU, and BERTScore. The Abstract and Table III report that fine-tuned Roberta-base achieves the highest scores (ROUGE-L 86.84%, BLEU 28.24%, BERTScore 95.38%), while the Conclusion states that Albert-base and Bert-base achieved the highest performance. The paper claims that fine-tuning substantially improves all models and that model capacity/architecture plays a major role.

Significance. If the empirical result is reproducible, the paper provides a useful, though incremental, benchmark comparison of fine-tuned transformer models on an extractive QA task. The main value would be as a practical guide to model selection for similar tasks. The paper does not introduce a new architecture, training method, or theoretical insight. It also does not provide code, data splits, or training hyperparameters, which limits its immediate utility and makes the central ranking difficult to verify. The internal inconsistency between the Conclusion and the Abstract/Table III is a serious correctness issue that must be resolved before the paper can be considered reliable.

major comments (4)
  1. [Section VI (Conclusion) vs. Abstract and Table III] The Abstract states that the fine-tuned Roberta-base model achieves the highest performance with ROUGE-L 86.84%, BLEU 28.24%, and BERTScore 95.38%, and Table III confirms these numbers are higher than Albert-base (82.27/26.56/93.85) and Bert-base (78.00/26.24/92.38). Section VI, however, says "models such as Albert-base and Bert-base achieved the highest performance." This is a direct contradiction about the paper's headline result. The reader cannot tell which statement is authoritative. Please correct the Conclusion or explain the discrepancy; if Table III is correct, the Conclusion is a factual error. This issue is load-bearing because the central claim is the model ranking.
  2. [Section IV-E and Section V-B] The fine-tuning procedure is described only as "the same training dataset and identical preprocessing steps" for all models. No learning rate, batch size, number of epochs, optimizer, warm-up, sequence length, or per-model hyperparameter search is reported. The models range from 4.4M (Bert-tiny) to 1.7B (Bloom), so a single fixed setup is not obviously fair across all architectures; the reported ranking (Roberta > Albert > Bert) could be an artifact of hyperparameter choice. Additionally, the train/validation/test split of SQuAD1.1 is not described, and it is unclear whether the reported numbers are on the official dev set or some other split. Without these details, the experiments are not reproducible and the ranking is not fully supported.
  3. [Section V (Results and Discussion)] All results are single-point estimates with no error bars, confidence intervals, or statistical significance tests. The differences between Roberta-base and Albert-base are small in relative terms (ROUGE-L 4.57 points, BLEU 1.68 points, BERTScore 1.53 points), and could fall within run-to-run variance. The paper claims "substantial improvements" and "best overall performance" without any variance analysis. Furthermore, SQuAD is conventionally evaluated with exact-match (EM) and F1; ROUGE-L and BLEU are not standard for extractive span QA, and BLEU is particularly unreliable for short answer spans. Please report EM/F1 or justify the chosen metrics and provide variance estimates.
  4. [Section IV-E, "five LLMs"] The text says "Each of the five LLMs is fine-tuned independently" but Tables II and III list twelve models. This discrepancy should be clarified: either the five LLMs are a subset and the remaining models are trained differently, or the sentence is a typo. As written, it is unclear whether all twelve models were fine-tuned under the same protocol.
minor comments (4)
  1. [Section III] The text refers to "SQuAD1.1 2.0" and "SQuAD1.1 1.1"; these should be "SQuAD2.0" and "SQuAD1.1". SQuAD1.1 contains only answerable questions; SQuAD2.0 adds unanswerable questions.
  2. [Figure 1] The text references Figure 1 as an overview of the system, but the figure is not included in the manuscript. Please ensure the figure is present or remove the reference.
  3. [Table I] The table heading "THESTANFORDQUESTIONANSWERINGDATASET" lacks spaces; should be "The Stanford Question Answering Dataset (SQuAD1.1)."
  4. [Section II] The sentence "Overall, Despite advances in LLM-based QA systems" has inconsistent capitalization; "Despite" should be lowercase. Minor typographical polish throughout would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the paper is an empirical fine-tuning and evaluation study; its internal Abstract/Conclusion inconsistency is a correctness/consistency issue, not a circularity issue.

full rationale

The paper's central claim is that fine-tuned Roberta-base achieves the best ROUGE-L/BLEU/BERTScore on SQuAD1.1. This is a measured empirical outcome, not a derived quantity that reduces to its inputs. The methodology defines a standard extractive QA loss (start/end cross-entropy), fine-tunes off-the-shelf models on a public benchmark, and evaluates with standard metrics. There is no fitted parameter that is later renamed as a prediction, no uniqueness theorem invoked from the authors' own prior work, and no ansatz smuggled in via citation. The several self-citations in the references are ordinary related-work citations and are not load-bearing for the experimental claim. The noted inconsistency between the Abstract/Table III (Roberta-base highest) and the Conclusion (Albert-base and Bert-base highest) is a serious internal consistency problem, but it is not circularity: neither claim is made true by construction. Likewise, the omitted hyperparameter details are a reproducibility concern, not a circular-derivation concern. The reader's assessment of 0.0 circularity is confirmed.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper relies on standard public resources (SQuAD, pretrained transformers) and does not introduce new theoretical constructs. The main unstated premises are that the chosen metrics are meaningful for extractive QA and that a single undisclosed training recipe is fair across heterogeneous models.

free parameters (2)
  • Fine-tuning hyperparameters (learning rate, batch size, epochs)
    Not disclosed in the paper. Scores likely depend on these and on whether each model received the same setting; no per-model search is described.
  • Evaluation split selection
    The paper refers to SQuAD training and validation sets but does not specify exactly which split produced the final numbers or whether any validation-based early stopping was used.
assumptions (4)
  • domain assumption SQuAD1.1 is a reliable benchmark for extractive QA
    Used as ground truth without discussing its limitations or potential biases; Section III.
  • domain assumption ROUGE-L, BLEU, and BERTScore are valid measures of answer accuracy
    Used as sole evaluation metrics; no exact-match/F1 or human evaluation is reported. This assumption is critical because the 'best model' ranking depends entirely on these metrics.
  • standard math Pretrained model implementations are correct and comparable
    Assumes that the HuggingFace or similar implementations used are bug-free and that the parameter counts are accurate; no versions or code are given.
  • ad hoc to paper One identical training setup is fair across very different model sizes and families
    Section IV-E states 'identical preprocessing steps' but gives no hyperparameters; equal epochs/learning rates likely disadvantage some models. The fairness of the comparison is assumed, not demonstrated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Answer Extraction in Context-based Question Answering Systems Using LLMs." pith.science (2026). https://pith.science/paper/KMKEIWOR

@misc{pith2026260606197,
  author       = {Pith},
  title        = {Pith review of: Improving Answer Extraction in Context-based Question Answering Systems Using LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KMKEIWOR}},
  note         = {Machine review of arXiv:2606.06197}
}
read the original abstract

Question answering (QA) systems have achieved notable progress with the advent of large language models (LLMs). However, they still face challenges in accurately extracting and generating precise answers from given contexts, particularly when dealing with complex or ambiguous queries. Existing approaches often struggle with contextual understanding, answer consistency, and generalization across diverse domains. In this work, we propose a question answering system based on large language models, where the input consists of a textual context and a corresponding question, and the output is a concise and accurate answer. The motivation behind this research lies in addressing the limitations of current QA systems, particularly their tendency to produce irrelevant or imprecise responses despite having access to the correct context. Our methodology involves fine-tuning a pre-trained LLM on a benchmark QA dataset to improve its contextual comprehension and answer extraction capabilities. Specifically, we utilize the Stanford Question Answering Dataset (SQuAD1.1), which provides high-quality context-question-answer triplets for supervised training and evaluation. Experimental results show that the fine-tuned Roberta-base model achieves the highest performance, attaining a ROUGE-L score of 86.84%, a BLEU score of 28.24%, and a BERTScore of 95.38%. These results indicate strong accuracy and answer relevance, demonstrating the effectiveness of the proposed approach for context-based question answering tasks. Furthermore, the findings confirm that targeted fine-tuning substantially improves the reliability and precision of QA systems.

Figures

Figures reproduced from arXiv: 2606.06197 by the authors.

Figure 1
Figure 1. Overview of the proposed context-based question answering system. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

Review history (3 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references

  1. [1]

    Nativqa: Multilingual culturally-aligned natural query for llms,

    M. A. Hasan, M. Hasanain, F. Ahmad, S. R. Laskar, S. Upadhyay, V . N. Sukhadia, M. Kutlu, S. A. Chowdhury, and F. Alam, “Nativqa: Multilingual culturally-aligned natural query for llms,” inFindings of the Association for Computational Linguistics: ACL 2025, 2025, pp. 14 886–14 909

    2025

  2. [2]

    You make me feel like a natural question: Training QA systems on transformed trivia questions,

    T. Kabir, Y . Y . Sung, S. Bandyopadhyay, H. Zou, A. Chandra, and J. L. Boyd-Graber, “You make me feel like a natural question: Training QA systems on transformed trivia questions,” inProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Y . Al-Onaizan, M. Bansal, and Y .-N. Chen, Eds. Miami, Florida, USA: Association for...

    2024

  3. [3]

    Exploring expected answer types for effective question answering systems for low resource language,

    C. Mallikarjuna and S. Sivanesan, “Exploring expected answer types for effective question answering systems for low resource language,” in Proceedings of the 21st International Conference on Natural Language Processing (ICON), S. Lalitha Devi and K. Arora, Eds. AU-KBC Research Centre, Chennai, India: NLP Association of India (NLPAI), Dec. 2024, pp. 12–20

    2024

  4. [4]

    LocalRQA: From generating data to lo- cally training, testing, and deploying retrieval-augmented QA systems,

    X. Yu, Y . Lu, and Z. Yu, “LocalRQA: From generating data to lo- cally training, testing, and deploying retrieval-augmented QA systems,” inProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), Y . Cao, Y . Feng, and D. Xiong, Eds. Bangkok, Thailand: Association for Computational Linguist...

    2024

  5. [5]

    How accurate are LLMs at multi- question answering on conversational transcripts?

    X. Zhu, S. Zong, and D. Rossouw, “How accurate are LLMs at multi- question answering on conversational transcripts?” inProceedings of the 2025 Conference on Empirical Methods in Natural Language Process- ing: Industry Track, S. Potdar, L. Rojas-Barahona, and S. Montella, Eds. Suzhou (China): Association for Computational Linguistics, Nov. 2025, pp. 1848–1855

    2025

  6. [6]

    Desiderata for the context use of question answering systems,

    S. Shaier, L. Hunter, and K. von der Wense, “Desiderata for the context use of question answering systems,” inProceedings of the 18th Con- ference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), Y . Graham and M. Purver, Eds. St. Julian’s, Malta: Association for Computational Linguistics, Mar. 2024, pp. 777–792

    2024

  7. [7]

    Context-aware an- swer extraction in question answering,

    Y . Seonwoo, J.-H. Kim, J.-W. Ha, and A. Oh, “Context-aware an- swer extraction in question answering,” inProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), B. Webber, T. Cohn, Y . He, and Y . Liu, Eds. Association for Computational Linguistics, Nov. 2020, pp. 2418–2428

    2020

  8. [8]

    Never lost in the middle: Mastering long-context question answering with position-agnostic decompositional training,

    J. He, K. Pan, X. Dong, Z. Song, L. LiuYiBo, Q. Qianguosun, Y . Liang, H. Wang, E. Zhang, and J. Zhang, “Never lost in the middle: Mastering long-context question answering with position-agnostic decompositional training,” inProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024, pp. 13 628–13 642

    2024

Show all 31 references
  1. [9]

    Augmenting compliance- guaranteed customer service chatbots: Context-aware knowledge expan- sion with large language models,

    M. Hong, C. J. Zhang, D. Jiang, and Y . He, “Augmenting compliance- guaranteed customer service chatbots: Context-aware knowledge expan- sion with large language models,” inProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track, ...

    2025

  2. [10]

    How proficient are large language models in formal languages? an in- depth insight for knowledge base question answering,

    J. Liu, S. Cao, J. Shi, T. Zhang, L. Nie, L. Hu, L. Hou, and J. Li, “How proficient are large language models in formal languages? an in- depth insight for knowledge base question answering,” inFindings of the Association for Computational Linguistics: ACL 2024, 2024, pp. 792–815

    2024

  3. [11]

    Webglm: Towards an efficient and reliable web-enhanced question-answering system,

    H. Lai, X. Liu, H. Yu, Y . Xu, I. L. Iong, S. Yao, A. Zeng, Z. Du, Y . Dong, and J. Tang, “Webglm: Towards an efficient and reliable web-enhanced question-answering system,”ACM Transactions on Information Systems, vol. 43, no. 5, pp. 1–43, 2025

    2025

  4. [12]

    Evaluating open- domain question answering in the era of large language models,

    E. Kamalloo, N. Dziri, C. Clarke, and D. Rafiei, “Evaluating open- domain question answering in the era of large language models,” in Proceedings of the 61st annual meeting of the Association for Compu- tational Linguistics (volume 1: long papers), 2023, pp. 5591–5606

    2023

  5. [13]

    Few-shot prompt- ing for extractive quranic qa with instruction-tuned llms,

    M. Basem, I. Oshallah, A. Hamdi, and A. Mohamed, “Few-shot prompt- ing for extractive quranic qa with instruction-tuned llms,” in2025 Intelligent Methods, Systems, and Applications (IMSA). IEEE, 2025, pp. 24–29

    2025

  6. [14]

    Two- stage quranic qa via ensemble retrieval and instruction-tuned answer extraction,

    M. Basem, I. Oshallah, A. Hamdi, K. Shaban, and H. Kassab, “Two- stage quranic qa via ensemble retrieval and instruction-tuned answer extraction,” in2025 IEEE/ACS 22nd International Conference on Com- puter Systems and Applications (AICCSA). IEEE, 2025, pp. 1–8

    2025

  7. [15]

    Cross-language approach for quranic qa,

    I. Oshallah, M. Basem, A. Hamdi, and A. Mohammed, “Cross-language approach for quranic qa,” inInternational Congress on Information and Communication Technology. Springer Nature Singapore Singapore, 2025, pp. 385–396

    2025

  8. [16]

    Hybrid graphs for table-and-text based question answering using llms,

    A. Agarwal, C. Devaguptapuet al., “Hybrid graphs for table-and-text based question answering using llms,” inProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long P...

    2025

  9. [17]

    Large language models meet knowledge graphs for question answering: Synthesis and opportunities,

    C. Ma, Y . Chen, T. Wu, A. Khan, and H. Wang, “Large language models meet knowledge graphs for question answering: Synthesis and opportunities,” inProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025, pp. 24 589–24 608

    2025

  10. [18]

    Aestar at semeval-2025 task 8: Agentic llms for question answering over tabular data,

    R. Tyagi, M. Gupta, and R. Bouri, “Aestar at semeval-2025 task 8: Agentic llms for question answering over tabular data,” inProceedings of the 19th International Workshop on Semantic Evaluation (SemEval- 2025), 2025, pp. 2249–2255

    2025

  11. [19]

    Opti- mized quran passage retrieval using an expanded qa dataset and fine- tuned language models,

    M. Basem, I. Oshallah, B. Hikal, A. Hamdi, and A. Mohamed, “Opti- mized quran passage retrieval using an expanded qa dataset and fine- tuned language models,” inThe International Conference of Advanced Computing and Informatics. Springer Nature Switzerland Cham, 2024, pp. 244–254

    2024

  12. [20]

    Trustuqa: A trustful framework for unified structured data question answering,

    W. Zhang, L. Jin, Y . Zhu, J. Chen, Z. Huang, J. Wang, Y . Hua, L. Liang, and H. Chen, “Trustuqa: A trustful framework for unified structured data question answering,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 24, 2025, pp. 25 931–25 939

    2025

  13. [21]

    Riro: Reshaping inputs, refining outputs unlocking the potential of large language models in data-scarce contexts,

    A. Hamdi, H. Kassab, M. Bahaa, and M. Mohamed, “Riro: Reshaping inputs, refining outputs unlocking the potential of large language models in data-scarce contexts,” inThe International Conference of Advanced Computing and Informatics. Springer Nature Switzerland Cham, 2024, pp. 69–79

    2024

  14. [22]

    Rationale-guided retrieval augmented generation for medical question answering,

    J. Sohn, Y . Park, C. Yoon, S. Park, H. Hwang, M. Sung, H. Kim, and J. Kang, “Rationale-guided retrieval augmented generation for medical question answering,” inProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Lingu...

    2025

  15. [23]

    Llm-daas: Llm-driven drone- as-a-service operations from text user requests,

    L. Wassim, K. Mohamed, and A. Hamdi, “Llm-daas: Llm-driven drone- as-a-service operations from text user requests,” inThe International Conference of Advanced Computing and Informatics. Springer, 2024, pp. 108–121

    2024

  16. [24]

    Few-shot optimized framework for hallucination detection in resource-limited nlp systems,

    B. Hikal, A. Nasreldin, A. Hamdi, and A. Mohammed, “Few-shot optimized framework for hallucination detection in resource-limited nlp systems,” inInternational Congress on Information and Communication Technology. Springer Nature Singapore Singapore, 2025, pp. 169–179

    2025

  17. [25]

    Sbu-nlp at semeval-2025 task 8: Self-correction and collaboration in llms for tabular question answer- ing,

    R. Rahnamoun and M. Shamsfard, “Sbu-nlp at semeval-2025 task 8: Self-correction and collaboration in llms for tabular question answer- ing,” inProceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), 2025, pp. 703–711

    2025

  18. [26]

    Coordinated llm multi-agent systems for collaborative question-answer generation,

    S. Saadaoui and E. Alonso, “Coordinated llm multi-agent systems for collaborative question-answer generation,”Knowledge-Based Systems, vol. 330, p. 114627, 2025

    2025

  19. [27]

    Large language model synergy for ensemble learning in medical question answering: design and evaluation study,

    H. Yang, M. Li, H. Zhou, Y . Xiao, Q. Fang, S. Zhou, and R. Zhang, “Large language model synergy for ensemble learning in medical question answering: design and evaluation study,”Journal of Medical Internet Research, vol. 27, p. e70080, 2025

    2025

  20. [28]

    Llm-sem: A sentiment- based student engagement metric using llms for e-learning platforms,

    A. Hamdi, A. A. Mazrou, and M. Shaltout, “Llm-sem: A sentiment- based student engagement metric using llms for e-learning platforms,” in The International Conference of Advanced Computing and Informatics. Springer, 2024, pp. 145–154

    2024

  21. [29]

    Llm-medqa: Enhancing medical question answering through case studies in large language models,

    H. Yang, H. Chen, H. Guo, Y . Chen, C.-S. Lin, S. Hu, J. Hu, X. Wu, and X. Wang, “Llm-medqa: Enhancing medical question answering through case studies in large language models,” in2025 International Joint Conference on Neural Networks (IJCNN). IEEE, 2025, pp. 1–8

    2025

  22. [30]

    Msa at semeval-2025 task 3: High quality weak labeling and llm ensemble verification for multilin- gual hallucination detection,

    B. Hikal, A. Nasreldin, and A. Hamdi, “Msa at semeval-2025 task 3: High quality weak labeling and llm ensemble verification for multilin- gual hallucination detection,” inProceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), 2025, pp. 989–995

    2025

  23. [31]

    Erpa: Efficient rpa model integrating ocr and llms for intelligent document processing,

    O. H. Abdellaif, A. N. Hassan, and A. Hamdi, “Erpa: Efficient rpa model integrating ocr and llms for intelligent document processing,” in2024 International Mobile, Intelligent, and Ubiquitous Computing Conference (MIUCC). IEEE, 2024, pp. 295–300

    2024

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.