Pith. sign in

REVIEW 5 major objections 5 minor 54 references

DragonVerseQA: Open-Domain Long-Form Context-Aware Question-Answering

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

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

Pith's one-line read DragonVerseQA is a new dataset that brings long-form question answering to serialized TV narratives.

desk verdict A potentially useful but unvalidated dataset paper whose evaluation metrics are internally inconsistent and whose 'first TV-QA' claim is contradicted by its own references. read the letter →

arxiv 2412.16694 v1 pith:QOTILCS5 submitted 2024-12-21 cs.CL cs.IR

Aritra Kumar Lahiri , Qinmin Vivian Hu This is my paper · ORCID
classification cs.CLcs.IR
keywords questionansweringdatasetlong-formQAnarrativeunderstandingTVseriescontext-awareknowledgegraphdatafilteringbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes DragonVerseQA, a question-answering dataset built for serialized TV narratives, centered on House of the Dragon and Game of Thrones. The authors argue that established fact-based QA benchmarks, which draw short answers mainly from encyclopedic articles, cannot capture character dynamics, plot progression, or audience interpretation. DragonVerseQA instead assembles a multi-source context from episode summaries, filtered user reviews, and structured character data, then generates long-form question-answer pairs from that context. If correct, it gives conversational AI and narrative-analysis research a domain-specific resource with substantially richer context and longer answers than existing open-domain long-form QA datasets.

What carries the argument

The load-bearing machinery is the context-aware question-answer generation pipeline. Each source chunk is first compressed by zero-shot abstractive summarization; then the BLANC block-attention model predicts answer spans, assigning soft context labels that decay with distance from the span; then an answer-aware question generator built on BERT-HLSQG, fine-tuned on ELI5, creates questions with [HL] highlight tokens bracketing the predicted span so duplicate mentions in long summaries do not mislead generation. A final automated stage screens each QA pair for spam and bias before it is added to the dataset.

What would settle it

Take a random sample of QA pairs that were not part of the 200-pair manual review and have two experts familiar with the series check whether each answer is grounded in its episode's source summaries and actually answers the question; if a meaningful fraction (say more than 10 percent) are unsupported, off-context, or hallucinated, the claim that DragonVerseQA is a high-quality context-aware benchmark would fail.

Watch

Extended reading notes

Core claim

The paper's central claim is that DragonVerseQA is the first QA dataset designed for the narrative complexity of serialized TV series, demonstrated on House of the Dragon Season 1 and Game of Thrones Season 8. It contains 3,200 long-form QA pairs, with questions averaging 15 words and answers averaging 31 words, each grounded in a multi-dimensional context built from episode summaries, filtered user reviews, and structured character metadata. The pairs are produced by a three-stage pipeline: zero-shot abstractive summarization of source chunks, answer-span prediction with the BLANC block-attention model, and answer-aware question generation using a BERT-HLSQG model fine-tuned on the ELI5 long-form QA dataset, followed by automated spam and bias filtering. The paper reports F1 of 85.6, BLEU of 0.45, ROUGE-L of 0.58, and a human-rated Context Relevance of 4.8 out of 5, and it presents a knowledge graph linking episodes, characters, and reviews to support narrative analysis.

Load-bearing premise

The quality of the 3,200 QA pairs rests on the automated generation pipeline working reliably across every episode, even though only 200 of those pairs were manually reviewed by human experts.

Editorial extensions

If this is right

  • QA systems evaluated on DragonVerseQA are tested on long, context-dependent answers rather than single factoid spans, which is closer to how people discuss serialized fiction.
  • The combination of episode summaries, reviews, and character metadata in one resource supports narrative understanding, sentiment analysis, and relation extraction without stitching together separate datasets.
  • The reported evaluation scores position DragonVerseQA as a stronger narrative-depth benchmark than established fact-based QA datasets, provided the quality check holds.
  • The knowledge graph linking episodes, characters, and reviews offers a structured view of narrative and audience response that can feed recommendation and personalization systems.

Reading between the lines

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

  • Editorial inference: if the automated pipeline generalizes, the same three-stage recipe could be reapplied to other serialized genres, turning DragonVerseQA into a template for domain-specific dataset construction.
  • Editorial inference: the main unverified risk is that only 200 of the 3,200 QA pairs were manually reviewed; a larger independent audit would test whether the reported Context Relevance score holds across the full dataset.
  • Editorial inference: because answers are generated from summarized source chunks rather than raw episode text, the dataset may carry summarization artifacts; comparing a sample of answers against the original episode summaries would reveal how much context is preserved.
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 / 5 minor

Summary. The paper introduces DragonVerseQA, a question-answering dataset for the TV series 'House of the Dragon' and 'Game of Thrones', assembled from episode summaries, user reviews, and structured WikiData, with QA pairs generated by an automated pipeline (GPT-3 zero-shot summarization, BLANC answer-span prediction, and BERT-HLSQG question generation fine-tuned on ELI5). The paper claims that this is the first QA dataset capturing the complexities of serialized TV narratives and that it constitutes a new benchmark for OTT content. It reports comparisons with SQuAD 2.0, TriviaQA, and Natural Questions, plus manual and automated validation and ablation studies of the question-generation component.

Significance. If the dataset were rigorously validated and the novelty claim were accurate, DragonVerseQA could be a useful resource for narrative QA, particularly because it integrates multiple context sources and includes a knowledge-graph visualization and a semi-supervised spam/bias filtering procedure. The paper is transparent about some limitations and makes a prototype publicly available. However, the evidence presented does not currently support the central claims: the main evaluation metrics lack gold references, the ablation results are internally inconsistent, the manual validation covers only a small fraction of the data without agreement or error analysis, and the 'first QA dataset for TV series narratives' claim is contradicted by the paper's own related-work discussion. The significance of the contribution is therefore not established as submitted.

major comments (5)
  1. [Section V.A, Table VI] The reported F1, BLEU, and ROUGE-L scores for DragonVerseQA are not meaningful without a defined gold-reference answer for each generated QA pair. BLEU and ROUGE require reference texts, and it is unclear what the F1 score is computed against. Because the answers were produced by the same pipeline that created the dataset, these numbers largely measure the generator's self-consistency rather than dataset quality. This is load-bearing because the 'new benchmark' claim depends on demonstrating the quality of the dataset, and the current evaluation does not do so.
  2. [Section V.C, Tables IX-XII] The ablation tables use inconsistent scales for the same baseline configuration. Tables IX-XI report the baseline as BLEU 85.2, ROUGE 88.1, and Exact Match 78.4, while Table XII reports the same hyperparameters (learning rate 1e-4, batch size 32) as BLEU 0.55, ROUGE 0.60, and Exact Match 0.70. No explanation is given for this roughly 100x scale difference, so the relative conclusions of the ablation study are not interpretable.
  3. [Section V.B, Table VII] The manual validation covers only 200 of 3,200 QA pairs, with no inter-annotator agreement score, no error analysis, and no per-item check that answers are grounded in the source episode summaries. The 'Context Relevance' score in Table VI is likewise reported without a rubric, annotation instructions, or annotator agreement. Given that the remainder of the dataset is generated automatically, this is insufficient evidence for the claim that the dataset is high-quality and contextually accurate.
  4. [Section I.B vs. Section II] The claim that DragonVerseQA is 'the first QA dataset to capture the intricate complexities of TV series narratives' is directly contradicted by the paper's own related-work section, which cites TVQA, MovieQA, DramaQA, NarrativeQA, and the authors' earlier GameofthronesQA. The novelty claim needs to be substantially revised or removed.
  5. [Section III.D-E] The automated generation pipeline uses zero-shot GPT-3 summarization and a question-generation model fine-tuned only on ELI5, a domain very different from serialized TV narratives. The paper provides no factuality verification of the generated answers against the original episode summaries or reviews. Because the dataset quality claim rests entirely on this pipeline operating reliably across 3,200 pairs, this is a load-bearing correctness risk that needs to be addressed with per-item grounding checks or a substantially larger human evaluation.
minor comments (5)
  1. [Section IV.A] There is a typo: 'Game of the Thrones' should be 'Game of Thrones'.
  2. [Section III.A, Algorithm 2] The text says the same procedure is used for extracting user reviews and the algorithm is shown as 'Collect Episode Summaries'; the algorithm labels should be clarified so that the review-collection pseudocode is distinct.
  3. [Table IV] The ELI5 answer length is listed as 'around 40 words', but DragonVerseQA's average answer length is reported as 31 words in Table V; this comparison needs a consistent definition of answer length.
  4. [Equation (3)] The PDF rendering of Equation (3) is broken across lines, and the text should explicitly define the window size and the role of the decay hyperparameter q before the equation.
  5. [References] Several citations do not match their context: reference [10] is attributed to 'Ramakrishna et al.' but the listed work is by Hebbar et al.; reference [24] is cited as the source of the nlgeval library but the title refers to a knowledge-graph QA generation paper. Please verify and correct the bibliography.

Circularity Check

2 steps flagged · score 6.0 of 10

DragonVerseQA's headline F1/BLEU/ROUGE scores are self-measurements: no gold references are defined, so Tables VI and IX-XII evaluate the generation pipeline against its own outputs.

  1. self definitional [Section V.A (Key Evaluation Metrics), Table VI; dataset generation in Section III.E.3]
    "BLEU examines how closely the LLM-generated answers align with reference answers, while ROUGE checks the overlap between answers and references."

    DragonVerseQA's answers are produced by the Section III.E pipeline ('we use e2e-qa task prefix to obtain the final output QA pair for each episode'). No independent gold reference answers are defined for the 3,200 pairs; the Section V.B manual review covers only 200 pairs and gives categorical ratings, not reference texts. Hence the F1/BLEU/ROUGE values in Table VI have no external reference to measure against, so the only candidate 'reference answers' are the generated answers themselves. The reported quality scores therefore reduce, by construction, to a similarity of the generator's own outputs to itself (or are undefined), and Table VI cannot independently support the 'new benchmark' claim.

  2. fitted input called prediction [Section V.C (Ablation Studies), Tables IX-XII; cf. Section III.E]
    "We employ our Answer-Aware Question Generation Model for our question generation process, which is fine-tuned using BERT-HLSQG [2]. In our ablation study, this model serves as the baseline..."

    The ablation evaluates the same Answer-Aware Question Generation model that generated DragonVerseQA's QA pairs (Section III.E.2-E.3). Tables IX-XI report BLEU/ROUGE/Exact Match against no described gold-standard reference split; the test items are the model's own generated questions and answers. The baseline scores are therefore self-measurements of the generator, and the encoder/decoder/hyperparameter comparisons measure how well variants reproduce the pipeline's own outputs, not how well the dataset supports independent QA. Table XII reports the same configuration as BLEU 0.55 instead of 85.2, an unexplained scale mismatch confirming these numbers are not anchored to a stable external reference.

full rationale

The dataset construction itself is not circular: it collects episode summaries, reviews, and structured data, then applies GPT-3 summarization, BLANC answer-span selection, and BERT-HLSQG question generation. The circularity enters at the validation stage, because the paper never defines an independent gold reference set for DragonVerseQA. Its answers are machine-generated, and its F1/BLEU/ROUGE scores are therefore either computed against those same generated answers or are undefined; either way, Table VI and the ablation tables do not provide external evidence of dataset quality. The only genuinely independent evidence is the small manual and MTurk review, which is limited to 200 pairs and lacks inter-annotator agreement details, so it cannot rescue the full 3,200-pair benchmark claim. The paper's Related Works also lists TVQA, MovieQA, DramaQA, NarrativeQA, FairytaleQA, and the authors' prior GameofthronesQA, which contradicts the 'first QA dataset for TV narratives' novelty claim; this is a novelty and correctness issue rather than a circular-reasoning issue, so it is noted but not scored here. Overall, one or more reported 'predictions' reduce by construction to self-measurement, while the underlying data resource retains some independent content, giving a partial circularity score of 6.

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

The dataset's quality claims rest on several unverified domain assumptions about the automated processing pipeline rather than on mathematical axioms. The free parameters are hand-chosen hyperparameters in the generation and filtering stages; none are fitted to an external benchmark, but their settings affect the final dataset composition.

free parameters (4)
  • q (BLANC decay rate) = not specified
    Hyperparameter in Equation 3 controlling the exponential decay of soft context labels for words far from the answer span. Chosen without stated justification; affects answer span prediction.
  • k (top-k reviews) = 10
    Number of high-quality IMDb and Rotten Tomatoes reviews kept per episode in the context. Chosen by hand and stated in Section I.
  • Sentiment polarity threshold = between -0.5 and 0.5
    Reviews with polarity scores outside this range are filtered out as biased. Stated in Section III.C without empirical justification.
  • Confidence threshold for pseudo-labeling = p > 0.9 or p < 0.1
    Threshold in Algorithm 3 for selecting confident pseudo-labels in the semi-supervised spam detection. Chosen by hand; no sensitivity analysis.
assumptions (4)
  • domain assumption GPT-3 zero-shot summarization produces faithful, contextually rich summaries of episode and review texts.
    The dataset context and subsequent QA generation depend on the summaries being accurate and preserving narrative details. No human verification of the summaries is reported.
  • domain assumption The pre-trained BERT spam detection model and the SVM bias model correctly classify spam and biased content.
    These filters are applied to all reviews and QA pairs, but the paper provides no evaluation of their accuracy or error rates.
  • domain assumption The fine-tuned BERT-HLSQG question generation model produces questions that are relevant to the predicted answer span.
    Automated QA generation is used for the full dataset; only 200 pairs are manually reviewed, so the quality of the remaining 3,000 pairs depends on this assumption.
  • domain assumption A random human review of 200 QA pairs is representative of the entire dataset.
    The paper generalizes quality scores from 200 pairs to the whole 3,200-pair dataset without reporting sampling variance or inter-annotator agreement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DragonVerseQA: Open-Domain Long-Form Context-Aware Question-Answering." pith.science (2026). https://pith.science/paper/QOTILCS5

@misc{pith2026241216694,
  author       = {Pith},
  title        = {Pith review of: DragonVerseQA: Open-Domain Long-Form Context-Aware Question-Answering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QOTILCS5}},
  note         = {Machine review of arXiv:2412.16694}
}
read the original abstract

This paper proposes a novel approach to develop an open-domain and long-form Over-The-Top (OTT) Question-Answering (QA) dataset, DragonVerseQA, specifically oriented to the fantasy universe of "House of the Dragon" and "Game Of Thrones" TV series. Most existing QA datasets focus on short, fact-based answers sourced almost solely from Wikipedia articles, devoid of depth and contextual richness for sophisticated narrative understanding. We curate a dataset that combines full episode summaries sourced from HBO and fandom wiki websites, user reviews from sources like IMDb and Rotten Tomatoes, and high-quality, open-domain, legally admissible sources, and structured data from repositories like WikiData into one dataset. The dataset provides a multi-dimensional context, reflecting complex character dynamics and plot developments from these varied sources. That means, on equal footing, only after heavy data preprocessing and filtering methods will meaningful, non-spam unbiased reviews be available in this enriched dataset. The comprehensive insights are given through the long-form answers generated from this enriched context. This is what makes this valuable dataset for improving conversational AI, narrative analysis, sentiment analysis, summarization techniques, and relation extraction. A comparative analysis with state-of-the-art QA datasets such as SQuAD 2.0, TriviaQA, and Natural Questions brings to light the unique advantages of our dataset in terms of contextual complexity and answer length. Detailed reviews add layers to audience sentiment and narrative interpretation, raising the bar for domain-specific QA with a new quality benchmark. Our work also allows a deeper understanding of entertainment-industry content and opens the door to more knowledgeable and creative AI-driven interactions within digital media environments.

Figures

Figures reproduced from arXiv: 2412.16694 by the authors.

Figure 1
Figure 1. DragonVerseQA - Dataset structure understanding of context. Integrating multi-dimensional context uniquely enables handling these narrative intrica￾cies within the dataset. • Long-Form Answers with Narrative Depth - We introduce long-form answers, capturing the nuances of narrative content. High-quality QA pairs allow users to grasp the storyline of each episode in context, offering a seamless experience by followin… view at source ↗
Figure 2
Figure 2. Filtered User Reviews [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Contextual Data Integration labels for the remaining unlabeled data of the user reviews by iterating the process to refine the model continually. Algorithm 3 describes the semi-supervised learning approach for the spam detection flow. The steps involved splitting the initial labeled data into feature vectors. After that, we train the initial spam detection model using ”RandomForestClassifier” and evaluate the metric… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Context-Aware Question Answering: Architecture [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: DragonVerseQA Sample QA pairs sample illustration of the generated long-form QA pairs from Episode 1 of the ”House of The Dragon” TV series. We also incorporate Spam Detection and Bias Filtering in our pipeline for each QA pair before appending it to the dataset. Algor…
Figure 8
Figure 8. Figure 8: Knowledge Graph Representation - DragonVerseQA [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 7
Figure 7. Figure 7: Factoid Questions Category Distribution [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 1
Figure 1. Figure 1: In assessing answer length appropriateness, the script [PITH_FULL_IMAGE:figures/full_fig_p008_1.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 38 canonical work pages

  1. [1]

    Devlin, J., Chang, M.W., Lee, K., & Toutanova, K. (2019). BERT: Pre- training of Deep Bidirectional Transformers for Language Understand- ing. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT)

    work page 2019

  2. [2]

    Ying-Hong Chan, Yao-Chung Fan. A Recurrent BERT-based Model for Question Generation, Proceedings of the Second Workshop on Machine Reading for Question Answering, pages 154–162 Hong Kong, China, November 4, 2019.(ACL)

    work page 2019

  3. [3]

    ”ELI5: Long-form question answering.” arXiv preprint arXiv:1907.09190 (2019)

    Fan, Angela, Yacine Jernite, Ethan Perez, David Grangier, Jason We- ston, and Michael Auli. ”ELI5: Long-form question answering.” arXiv preprint arXiv:1907.09190 (2019)

    arXiv 2019

  4. [4]

    Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang

  5. [5]

    Weld, and Luke Zettlemoyer

    Joshi, Mandar, Eunsol Choi, Daniel S. Weld, and Luke Zettlemoyer. ”Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension.” arXiv preprint arXiv:1705.03551 (2017)

    arXiv 2017

  6. [6]

    ”Natural questions: a benchmark for question answering research.” Transactions of the Association for Computational Linguistics 7 (2019): 453-466

    Kwiatkowski, Tom, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein et al. ”Natural questions: a benchmark for question answering research.” Transactions of the Association for Computational Linguistics 7 (2019): 453-466. 10

    work page 2019

  7. [7]

    ”Movieqa: Understanding stories in movies through question-answering.” In Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Tapaswi, Makarand, Yukun Zhu, Rainer Stiefelhagen, Antonio Torralba, Raquel Urtasun, and Sanja Fidler. ”Movieqa: Understanding stories in movies through question-answering.” In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4631-4640. 2016

    work page 2016

  8. [8]

    Lei, Jie, Licheng Yu, Mohit Bansal, and Tamara L. Berg. ”Tvqa: Localized, compositional video question answering.” arXiv preprint arXiv:1809.01696 (2018)

    arXiv 2018

Show all 54 references
  1. [9]

    ”The netflix recommender system: Algorithms, business value, and innovation.” ACM Transactions on Management Information Systems (TMIS) 6, no

    Gomez-Uribe, Carlos A., and Neil Hunt. ”The netflix recommender system: Algorithms, business value, and innovation.” ACM Transactions on Management Information Systems (TMIS) 6, no. 4 (2015): 1-19

    2015

  2. [10]

    Hebbar, Rajat, Krishna Somandepalli, and Shrikanth Narayanan. ”Ro- bust speech activity detection in movie audio: Data resources and experimental evaluation.” In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 4105-4109. ...

    2019

  3. [11]

    Bender, Amandalynne Paullada, Emily Denton, and Alex Hanna

    Raji, Inioluwa Deborah, Emily M. Bender, Amandalynne Paullada, Emily Denton, and Alex Hanna. ”AI and the everything in the whole wide world benchmark.” arXiv preprint arXiv:2111.15366 (2021)

    2021 arXiv

  4. [12]

    ”The narra- tiveqa reading comprehension challenge.” Transactions of the Associa- tion for Computational Linguistics 6 (2018): 317-328

    Ko ˇcisk´y, Tom ´aˇs, Jonathan Schwarz, Phil Blunsom, Chris Dyer, Karl Moritz Hermann, G ´abor Melis, and Edward Grefenstette. ”The narra- tiveqa reading comprehension challenge.” Transactions of the Associa- tion for Computational Linguistics 6 (2018): 317-328

    2018

  5. [13]

    ”Ms marco: A human-generated machine reading comprehension dataset.” (2016)

    Nguyen, Tri, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. ”Ms marco: A human-generated machine reading comprehension dataset.” (2016)

    2016

  6. [14]

    Cohen, Ruslan Salakhutdinov, and Christopher D

    Yang, Zhilin, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W. Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. ”HotpotQA: A dataset for diverse, explainable multi-hop question answering.” arXiv preprint arXiv:1809.09600 (2018)

    2018 arXiv

  7. [15]

    ”QuAC: Question answering in context.” arXiv preprint arXiv:1808.07036 (2018)

    Choi, Eunsol, He He, Mohit Iyyer, Mark Yatskar, Wen-tau Yih, Yejin Choi, Percy Liang, and Luke Zettlemoyer. ”QuAC: Question answering in context.” arXiv preprint arXiv:1808.07036 (2018)

    2018 arXiv

  8. [16]

    Reddy, Siva, Danqi Chen, and Christopher D. Manning. ”Coqa: A conversational question answering challenge.” Transactions of the As- sociation for Computational Linguistics 7 (2019): 249-266

    2019

  9. [17]

    ”Wikiqa: A challenge dataset for open-domain question answering.” In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp

    Yang, Yi, Wen-tau Yih, and Christopher Meek. ”Wikiqa: A challenge dataset for open-domain question answering.” In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pp. 2013-2018. 2015

    2015

  10. [18]

    https://beautiful-soup-4.readthedocs.io/en/latest/

  11. [19]

    https://huggingface.co/mrm8488/bert-tiny-finetuned-sms-spam-detection

  12. [20]

    https://textblob.readthedocs.io/en/dev/

  13. [21]

    ”Context- aware answer extraction in question answering.” arXiv preprint arXiv:2011.02687 (2020)

    Seonwoo, Yeon, Ji-Hoon Kim, Jung-Woo Ha, and Alice Oh. ”Context- aware answer extraction in question answering.” arXiv preprint arXiv:2011.02687 (2020)

    2020 arXiv

  14. [22]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu

  15. [23]

    ”Rouge: A package for automatic evaluation of sum- maries.” Text summarization branches out

    Lin, Chin-Yew. ”Rouge: A package for automatic evaluation of sum- maries.” Text summarization branches out. 2004

    2004

  16. [24]

    Indurthi, Sathish Reddy, et al. ”Generating natural language question- answer pairs from a knowledge graph using an based question generation model.” Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: V olume 1, Long Pa...

    2017

  17. [25]

    (2017, December)

    Hui, Z., Liu, X., & Sun, M. (2017, December). A Context-aware Attention Network for Interactive Question Answering. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers) (pp. 1212-1222)

    2017

  18. [26]

    Barla, et al

    Cambazoglu, B. Barla, et al. ”A Review of Public Datasets in Question Answering Research.”

  19. [27]

    ”Exploring the limits of transfer learning with a unified text-to-text transformer.” arXiv preprint arXiv:1910.10683 (2019)

    Raffel, Colin, et al. ”Exploring the limits of transfer learning with a unified text-to-text transformer.” arXiv preprint arXiv:1910.10683 (2019)

    2019 arXiv

  20. [28]

    ”TWEETQA: A social media focused question answering dataset.” arXiv preprint arXiv:1907.06292 (2019)

    Xiong, Wenhan, et al. ”TWEETQA: A social media focused question answering dataset.” arXiv preprint arXiv:1907.06292 (2019)

    2019 arXiv

  21. [29]

    ”Gameofthronesqa: Answer-aware question-answer pairs for tv series.” In European Con- ference on Information Retrieval, pp

    Lahiri, Aritra Kumar, and Qinmin Vivian Hu. ”Gameofthronesqa: Answer-aware question-answer pairs for tv series.” In European Con- ference on Information Retrieval, pp. 180-189. Cham: Springer Interna- tional Publishing, 2022

    2022

  22. [30]

    ”Benchmarking large language models in retrieval-augmented generation.” In Proceed- ings of the AAAI Conference on Artificial Intelligence, vol

    Chen, Jiawei, Hongyu Lin, Xianpei Han, and Le Sun. ”Benchmarking large language models in retrieval-augmented generation.” In Proceed- ings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 16, pp. 17754-17762. 2024

    2024

  23. [31]

    Xinya Du and Claire Cardie. 2018. Harvesting Paragraph-level Question- Answer Pairs from Wikipedia. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers). 1907–1917

    2018

  24. [32]

    Sanh, V . (2019). DistilBERT, A Distilled Version of BERT: Smaller, Faster, Cheaper, and Lighter. arXiv preprint arXiv:1910.01108

    2019 arXiv

  25. [33]

    Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language Models are Unsupervised Multitask Learn- ers—OpenAI blog

    2019

  26. [34]

    Chen, Q., Zhu, X., Ling, Z.H., Wei, S., & Jiang, H. (2019). Context- aware question answering over dialogue with attention-based LSTM net- works. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferenc...

    2019

  27. [35]

    Brownlee, J., Thorne, J., Houlsby, N., Groot, R., & Clark, P. (2019). Question Answering with Neural Attention-based Hierarchical Contexts. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on N...

    2019

  28. [36]

    Lopez, Luis Enrico, et al. ”Simplifying paragraph-level question gen- eration via transformer language models.” PRICAI 2021: Trends in Artificial Intelligence: 18th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2021, Hanoi, Vietnam, November 8–12, 2021

    2021

  29. [37]

    Tong Wang, Xingdi Yuan, and Adam Trischler. 2017. A joint model for question answering and question generation. arXiv preprint arXiv:1706.01450 (2017)

    2017 arXiv

  30. [38]

    Nan Duan, Duyu Tang, Peng Chen, and Ming Zhou. 2017. Question generation for question answering. In Proceedings of the 2017 Con- ference on Empirical Methods in Natural Language Processing, pages 866–874.. http://www.aclweb.org/anthology/D13-1160

    2017

  31. [39]

    Sandeep Subramanian, Tong Wang, Xingdi Yuan, Saizheng Zhang, Adam Trischler, and Yoshua Bengio. 2018. Neural Models for Key Phrase Extraction and Question Generation. In Proceedings of the Workshop on Machine Reading for Question Answering. 78–88

    2018

  32. [40]

    Liangming Pan, Wenqiang Lei, Tat-Seng Chua, and Min-Yen Kan

  33. [41]

    Xingwu Sun, Jing Liu, Yajuan Lyu, Wei He, Yanjun Ma, and Shi Wang

  34. [42]

    ”Onestop qamaker: extract question-answer pairs from text in a one-stop approach.” arXiv preprint arXiv:2102.12128 (2021)

    Cui, Shaobo, et al. ”Onestop qamaker: extract question-answer pairs from text in a one-stop approach.” arXiv preprint arXiv:2102.12128 (2021)

    2021 arXiv

  35. [43]

    Open domain question answering using Wikipedia-based knowledge model

    Pum-Mo Ryu, Myung-Gil Jang, and Hyun-Ki Kim.2014. Open domain question answering using Wikipedia-based knowledge model. Informa- tion Processing and Management, 50(5):683 – 692

    2014

  36. [44]

    ”Narrative question answering with cutting-edge open-domain qa techniques: A comprehensive study.” Transactions of the Association for Computational Linguistics 9 (2021): 1032-1046

    Mou, Xiangyang, Chenghao Yang, Mo Yu, Bingsheng Yao, Xiaox- iao Guo, Saloni Potdar, and Hui Su. ”Narrative question answering with cutting-edge open-domain qa techniques: A comprehensive study.” Transactions of the Association for Computational Linguistics 9 (2021): 1032-1046

    2021

  37. [45]

    ”Evaluating question answering evaluation.” In Proceedings of the 2nd workshop on machine reading for question answering, pp

    Chen, Anthony, Gabriel Stanovsky, Sameer Singh, and Matt Gardner. ”Evaluating question answering evaluation.” In Proceedings of the 2nd workshop on machine reading for question answering, pp. 119-124. 2019

    2019

  38. [46]

    ”Dramaqa: Character-centered video story under- standing with hierarchical qa.” Proceedings of the AAAI Conference on Artificial Intelligence

    Choi, Seongho, et al. ”Dramaqa: Character-centered video story under- standing with hierarchical qa.” Proceedings of the AAAI Conference on Artificial Intelligence. V ol. 35. No. 2. 2021

    2021

  39. [47]

    K., Chambers, N., Mooney, R., & Balasubramanian, N

    Lal, Y . K., Chambers, N., Mooney, R., & Balasubramanian, N. (2021). TellMeWhy: A dataset for answering why-questions in narratives. arXiv preprint arXiv:2106.06132

    2021 arXiv

  40. [48]

    & Warschauer, M

    Xu, Y ., Wang, D., Yu, M., Ritchie, D., Yao, B., Wu, T., ... & Warschauer, M. (2022). Fantastic Questions and Where to Find Them: FairytaleQA– An Authentic Dataset for Narrative Comprehension. arXiv preprint arXiv:2203.13947

    2022 arXiv

  41. [49]

    ”Distilling knowledge from reader to retriever for question answering.” arXiv preprint arXiv:2012.04584 (2020)

    Izacard, Gautier, and Edouard Grave. ”Distilling knowledge from reader to retriever for question answering.” arXiv preprint arXiv:2012.04584 (2020)

    2020 arXiv

  42. [50]

    ”FeTaQA: Free-form table question answering.” Transactions of the Association for Computational Linguistics 10 (2022): 35-49

    Nan, Linyong, Chiachun Hsieh, Ziming Mao, Xi Victoria Lin, Neha Verma, Rui Zhang, Wojciech Kry´sci´nski et al. ”FeTaQA: Free-form table question answering.” Transactions of the Association for Computational Linguistics 10 (2022): 35-49

    2022

  43. [2002]

    In Proceedings of 40th Annual Meeting of the Asso- ciation for Computational Linguistics

    Bleu: a method for automatic evaluation of machine trans- lation. In Proceedings of 40th Annual Meeting of the Asso- ciation for Computational Linguistics. Association for Computa- tional Linguistics, Philadelphia, Pennsylvania, USA, pages 311–318. https://doi.org/10.3115/1073...

  44. [2016]

    In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP)

    Squad: 100,000+ questions for machine comprehension of text. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP). ACL, Austin, Texas, pages 2383–2392

    2016

  45. [2018]

    In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing

    Answer-focused and position-aware neural question generation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. 3930–3939

    2018

  46. [2019]

    arXiv preprint arXiv:1905.08949 (2019)

    Recent advances in neural question generation. arXiv preprint arXiv:1905.08949 (2019)

    2019 arXiv

Pith tools

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