arxiv: 2604.19685 · v1 · submitted 2026-04-21 · 💻 cs.CL

Recognition: unknown

An Answer is just the Start: Related Insight Generation for Open-Ended Document-Grounded QA

Aparna Garimella, Koyel Mukherjee, Pritika Ramu, Saransh Sharma

Authors on Pith no claims yet

Pith reviewed 2026-05-10 02:42 UTC · model grok-4.3

classification 💻 cs.CL

keywords document-grounded QAopen-ended question answeringrelated insight generationthematic clusteringgraph neighborhood selectionSCOpE-QA datasetLLM insight generation

0 comments

The pith

InsightGen generates related insights from document collections to help refine answers to open-ended questions.

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

The paper introduces document-grounded related insight generation as a new task to support users who refine open-ended questions through multiple iterations rather than stopping at one answer. It releases the SCOpE-QA dataset of 3,000 questions drawn from 20 research collections to enable work on this task. InsightGen addresses the task with a two-stage process that first clusters documents into themes and then selects neighboring contexts from the resulting graph to feed into LLMs for insight creation. Evaluation across 3,000 questions, two generation models, and two settings finds the outputs useful, relevant, and actionable, providing a baseline for future systems.

Core claim

By first constructing a thematic representation of the document collection using clustering and then selecting related context based on neighborhood selection from the thematic graph, InsightGen enables LLMs to produce diverse and relevant insights that improve, extend, or rethink an initial answer to an open-ended question.

What carries the argument

InsightGen's two-stage pipeline of thematic clustering of documents followed by graph neighborhood selection to retrieve context for LLM-based insight generation.

If this is right

Open-ended QA can shift from single factual responses to supporting iterative refinement with document-derived extensions.
The SCOpE-QA dataset provides a concrete benchmark for measuring insight usefulness and relevance in scientific collections.
Thematic graph methods allow LLMs to access diverse contexts that would otherwise require manual user navigation.
Generated insights remain actionable across different underlying generation models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Embedding this style of insight generation into existing search tools could shorten the number of rounds users need to reach a satisfying answer.
The clustering-plus-graph approach might transfer to non-research domains such as news archives or policy documents where open-ended questions also arise.
If the thematic step proves essential, future work could test whether simpler similarity-based retrieval produces comparable diversity.

Load-bearing premise

Clustering documents into themes and selecting graph neighborhoods will reliably surface context that leads to useful and diverse insights for open-ended questions.

What would settle it

If human raters judge insights from InsightGen no more useful or relevant than those produced by direct LLM prompting on the full document collection without clustering or graph steps, the value of the two-stage machinery would be undermined.

Figures

Figures reproduced from arXiv: 2604.19685 by Aparna Garimella, Koyel Mukherjee, Pritika Ramu, Saransh Sharma.

**Figure 2.** Figure 2: SCOpE-QA dataset curation pipeline overview, highlighting collection curation and QA generation. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: INSIGHTGEN pipeline showing theme-based clustering, context selection, and insight generation creative exploration, including identifying missing information, proposing new ideas, suggesting alternate answer framings, creating mind maps, highlighting potential issues or objections, presenting interesting facts, designing short quizzes, providing real-world applications or analogies, and analyzing tradeo… view at source ↗

**Figure 4.** Figure 4: Example insights showing differences in type and content. These insights provide additional observations [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗

**Figure 5.** Figure 5: Example insights showing differences in type and content. These insights provide additional observations [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗

**Figure 6.** Figure 6: Thematic clusters from the Graph ML collection. Each node corresponds to a document segment, and [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗

**Figure 7.** Figure 7: Clusters within the RL collection. Nodes represent semantically grouped document segments, showing [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗

**Figure 8.** Figure 8: Clusters in the Quantization collection. Nodes correspond to document segments, and edges indicate [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗

read the original abstract

Answering open-ended questions remains challenging for AI systems because it requires synthesis, judgment, and exploration beyond factual retrieval, and users often refine answers through multiple iterations rather than accepting a single response. Existing QA benchmarks do not explicitly support this refinement process. To address this gap, we introduce a new task, document-grounded related insight generation, where the goal is to generate additional insights from a document collection that help improve, extend, or rethink an initial answer to an open-ended question, ultimately supporting richer user interaction and a better overall question answering experience. We curate and release SCOpE-QA (Scientific Collections for Open-Ended QA), a dataset of 3,000 open-ended questions across 20 research collections. We present InsightGen, a two-stage approach that first constructs a thematic representation of the document collection using clustering, and then selects related context based on neighborhood selection from the thematic graph to generate diverse and relevant insights using LLMs. Extensive evaluation on 3,000 questions using two generation models and two evaluation settings shows that InsightGen consistently produces useful, relevant, and actionable insights, establishing a strong baseline for this new task.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper defines a new task for generating related insights after an initial open-ended QA answer and releases the SCOpE-QA dataset, but the evaluation provides no numbers, baselines, or details on how usefulness was judged.

read the letter

The main takeaway is that this work formalizes document-grounded related insight generation as a distinct task and ships a new dataset of 3000 questions across 20 collections. InsightGen clusters documents into themes, builds a graph, and selects neighborhoods to prompt LLMs for follow-up insights that extend or refine a starting answer. That framing targets a practical user behavior that standard QA benchmarks overlook, and the data release gives the community something concrete to build on. Releasing SCOpE-QA is the clearest positive step here; new collections focused on open-ended scientific questions are still scarce. The two-stage pipeline is a reasonable heuristic attempt to surface complementary context rather than just more facts. The soft spots are in the results. The abstract claims consistent production of useful, relevant, and actionable insights across models and settings, yet supplies no quantitative scores, no comparison to simpler retrieval baselines, and no account of how those three qualities were measured or validated. Without those controls it is difficult to know whether the clustering-plus-graph step actually adds value or whether random additional passages would perform similarly. The stress-test concern about neighborhood selection pulling near-duplicates instead of fresh perspectives therefore remains open. This paper is aimed at researchers working on iterative or exploratory QA systems and on LLM pipelines that support refinement loops. Readers who need a starting benchmark for follow-up insight tasks will find the dataset and task definition worth examining. It deserves a serious referee because the task definition and data release are substantive enough to merit external feedback, even though the current experiments need tightening. I would recommend sending it to review with explicit requests for baseline comparisons and clearer metric definitions.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces the task of document-grounded related insight generation to support iterative refinement of answers to open-ended questions. It releases the SCOpE-QA dataset of 3,000 questions across 20 scientific collections and proposes InsightGen, a two-stage pipeline that first performs thematic clustering of the document collection and then selects context via neighborhood selection in the resulting graph to prompt LLMs for additional insights. The authors claim that evaluations using two generation models and two settings on the full dataset demonstrate that InsightGen produces useful, relevant, and actionable insights, thereby establishing a strong baseline for the new task.

Significance. If the evaluation claims hold with proper controls, the work would be significant for shifting QA research from single-answer retrieval toward interactive, exploratory systems, especially in scientific domains where synthesis and multiple perspectives matter. The public release of SCOpE-QA is a clear positive that could seed follow-on benchmarks. However, the absence of quantitative metrics and baselines in the reported results substantially reduces the assessed significance at present.

major comments (2)

[Abstract] Abstract: the claim that 'InsightGen consistently produces useful, relevant, and actionable insights' and 'establishes a strong baseline' rests on evaluations across 3,000 questions, yet the abstract (and by extension the reported results) supplies no numerical scores, inter-annotator agreement, or baseline comparisons (e.g., against direct passage sampling or random neighborhood selection). This directly undermines the central empirical claim.
[Method and Evaluation] Method and Evaluation sections: the two-stage pipeline (thematic clustering followed by graph neighborhood selection) is presented as the key mechanism for surfacing complementary context, but no ablation or comparison is described against a simpler retrieval baseline that simply samples additional passages. Without such a control, it remains unverified whether the clustering and graph steps reliably improve insight quality or merely retrieve near-duplicates, which is load-bearing for the assertion that the method works 'consistently'.

minor comments (2)

[Dataset] Dataset curation: provide more detail on how the 20 collections were chosen, the criteria for question creation, and whether the collections themselves are released alongside the questions.
[Method] Notation: the description of the 'thematic graph' would benefit from an explicit definition or pseudocode for neighborhood selection to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our work introducing the related insight generation task and the InsightGen approach. We address each major comment below and commit to revisions that will strengthen the empirical support and clarity of the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that 'InsightGen consistently produces useful, relevant, and actionable insights' and 'establishes a strong baseline' rests on evaluations across 3,000 questions, yet the abstract (and by extension the reported results) supplies no numerical scores, inter-annotator agreement, or baseline comparisons (e.g., against direct passage sampling or random neighborhood selection). This directly undermines the central empirical claim.

Authors: We agree that the abstract would be strengthened by including key quantitative results to support the claims of usefulness, relevance, and actionability. The Evaluation section reports human assessment results across two models and two settings on the full SCOpE-QA dataset, including average ratings for the three criteria and inter-annotator agreement statistics. In the revised version, we will update the abstract to concisely report these numerical scores (e.g., mean ratings and agreement levels) along with a brief note on the evaluation design. This change will make the central empirical claim more transparent while preserving the abstract's brevity. revision: yes
Referee: [Method and Evaluation] Method and Evaluation sections: the two-stage pipeline (thematic clustering followed by graph neighborhood selection) is presented as the key mechanism for surfacing complementary context, but no ablation or comparison is described against a simpler retrieval baseline that simply samples additional passages. Without such a control, it remains unverified whether the clustering and graph steps reliably improve insight quality or merely retrieve near-duplicates, which is load-bearing for the assertion that the method works 'consistently'.

Authors: We acknowledge that an explicit ablation against a simpler passage-sampling baseline is necessary to isolate the contribution of the thematic clustering and graph neighborhood steps. The design rationale is that direct sampling (e.g., random or similarity-based passage retrieval) often yields near-duplicates or less thematically coherent context, whereas the graph neighborhood approach aims to select diverse yet related insights. To address this, we will add a controlled comparison in the revised Evaluation section: InsightGen versus a baseline that samples additional passages using embedding similarity to the question and initial answer. We will report the same human evaluation metrics for both, allowing readers to assess whether the two-stage pipeline yields measurable gains in insight quality. revision: yes

Circularity Check

0 steps flagged

No circularity: heuristic pipeline evaluated empirically on new dataset

full rationale

The paper introduces a new task and a two-stage heuristic method (thematic clustering followed by graph neighborhood selection to retrieve context for LLM-based insight generation). No equations, fitted parameters, or derivations are present. The central claim rests on empirical evaluation across 3,000 questions using two models and two settings on the newly curated SCOpE-QA dataset. No self-citations are load-bearing for any uniqueness theorem or ansatz, and no predictions reduce to inputs by construction. The method is self-contained as an empirical baseline without circular reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work is entirely empirical and relies on standard NLP assumptions about clustering and graph neighborhoods rather than new mathematical constructs or fitted parameters.

axioms (2)

domain assumption Thematic clustering of a document collection produces a useful representation for selecting related context.
Invoked in the first stage of InsightGen to build the thematic graph.
domain assumption Neighborhood selection on the thematic graph identifies context that yields relevant and diverse insights.
Core mechanism of the second stage before LLM generation.

pith-pipeline@v0.9.0 · 5513 in / 1300 out tokens · 37866 ms · 2026-05-10T02:42:07.087384+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

57 extracted references · 14 canonical work pages

[1]

2025 , eprint=

Sufficient Context: A New Lens on Retrieval Augmented Generation Systems , author=. 2025 , eprint=

2025
[2]

2024 , eprint=

LLMs Know What They Need: Leveraging a Missing Information Guided Framework to Empower Retrieval-Augmented Generation , author=. 2024 , eprint=

2024
[3]

2018 , eprint=

HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering , author=. 2018 , eprint=

2018
[4]

2020 , eprint=

Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Steps , author=. 2020 , eprint=

2020
[5]

2022 , eprint=

MuSiQue: Multihop Questions via Single-hop Question Composition , author=. 2022 , eprint=

2022
[6]

2024 , eprint=

I Could've Asked That: Reformulating Unanswerable Questions , author=. 2024 , eprint=

2024
[7]

Know What You Don 't Know : Unanswerable Questions for SQuAD

Rajpurkar, Pranav and Jia, Robin and Liang, Percy. Know What You Don ' t Know: Unanswerable Questions for SQ u AD. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). 2018. doi:10.18653/v1/P18-2124. 1806.03822 , archivePrefix=

work page doi:10.18653/v1/p18-2124 2018
[8]

2024 , eprint=

Clustered Retrieved Augmented Generation (CRAG) , author=. 2024 , eprint=

2024
[9]

2025 , eprint=

Efficient Dynamic Clustering-Based Document Compression for Retrieval-Augmented-Generation , author=. 2025 , eprint=

2025
[10]

2025 , eprint=

More Documents, Same Length: Isolating the Challenge of Multiple Documents in RAG , author=. 2025 , eprint=

2025
[11]

2024 , eprint=

Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG , author=. 2024 , eprint=

2024
[12]

ClusterChat: Multi-Feature Search for Corpus Exploration , url=

Chouhan, Ashish and Mandour, Saifeldin and Gertz, Michael , year=. ClusterChat: Multi-Feature Search for Corpus Exploration , url=. doi:10.1145/3726302.3730137 , booktitle=

work page doi:10.1145/3726302.3730137
[13]

2025 , eprint=

SAKR: Enhancing Retrieval-Augmented Generation via Streaming Algorithm and K-Means Clustering , author=. 2025 , eprint=

2025
[14]

Towards Answer-unaware Conversational Question Generation

Nakanishi, Mao and Kobayashi, Tetsunori and Hayashi, Yoshihiko. Towards Answer-unaware Conversational Question Generation. Proceedings of the 2nd Workshop on Machine Reading for Question Answering. 2019. doi:10.18653/v1/D19-5809

work page doi:10.18653/v1/d19-5809 2019
[15]

Modeling What-to-ask and How-to-ask for Answer-unaware Conversational Question Generation

Do, Xuan Long and Zou, Bowei and Joty, Shafiq and Tai, Tran and Pan, Liangming and Chen, Nancy and Aw, Ai Ti. Modeling What-to-ask and How-to-ask for Answer-unaware Conversational Question Generation. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023. doi:10.18653/v1/2023.acl-long.603

work page doi:10.18653/v1/2023.acl-long.603 2023
[16]

Reinforced Dynamic Reasoning for Conversational Question Generation

Pan, Boyuan and Li, Hao and Yao, Ziyu and Cai, Deng and Sun, Huan. Reinforced Dynamic Reasoning for Conversational Question Generation. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019. doi:10.18653/v1/P19-1203

work page doi:10.18653/v1/p19-1203 2019
[17]

C hain CQG : Flow-Aware Conversational Question Generation

Gu, Jing and Mirshekari, Mostafa and Yu, Zhou and Sisto, Aaron. C hain CQG : Flow-Aware Conversational Question Generation. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021. doi:10.18653/v1/2021.eacl-main.177

work page doi:10.18653/v1/2021.eacl-main.177 2021
[18]

Conversational QA Dataset Generation with Answer Revision

Hwang, Seonjeong and Lee, Gary Geunbae. Conversational QA Dataset Generation with Answer Revision. Proceedings of the 29th International Conference on Computational Linguistics. 2022

2022
[19]

2024 , eprint=

Question Generation in Knowledge-Driven Dialog: Explainability and Evaluation , author=. 2024 , eprint=

2024
[20]

Proceedings of the 20th International Joint Conference on Artifical Intelligence , pages =

Pan, Shimei and Shaw, James , title =. Proceedings of the 20th International Joint Conference on Artifical Intelligence , pages =. 2007 , publisher =

2007
[21]

Companion Proceedings of the ACM Web Conference 2024 , pages =

Tayal, Anuja and Tyagi, Aman , title =. Companion Proceedings of the ACM Web Conference 2024 , pages =. 2024 , isbn =. doi:10.1145/3589335.3651905 , abstract =

work page doi:10.1145/3589335.3651905 2024
[22]

and Lau, Angela and Huang, Lifu and Sun, Tong

Lin, Zihao and Wang, Zichao and Pan, Yuanting and Manjunatha, Varun and Rossi, Ryan A. and Lau, Angela and Huang, Lifu and Sun, Tong. Persona- SQ : A Personalized Suggested Question Generation Framework For Real-world Documents. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Huma...

work page doi:10.18653/v1/2025.naacl-demo.20 2025
[23]

Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence,

Jiwei Tan and Xiaojun Wan and Jianguo Xiao , title =. Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence,. 2017 , doi =

2017
[24]

Scientific Reports , volume=

A scientific-article key-insight extraction system based on multi-actor of fine-tuned open-source large language models , author=. Scientific Reports , volume=. 2025 , publisher=

2025
[25]

2023 , howpublished =

OpenAI , title =. 2023 , howpublished =

2023
[26]

2023 , howpublished =

Google DeepMind , title =. 2023 , howpublished =

2023
[27]

2022 , howpublished =

Bai, Yuntao and Kadavath, Saurav and Kundu, Sandipan and others , title =. 2022 , howpublished =

2022
[28]

2023 , howpublished =

Touvron, Hugo and Lavril, Thibaut and Izacard, Gautier and others , title =. 2023 , howpublished =

2023
[29]

2025 , eprint=

Retrieval-Augmented Generation: A Comprehensive Survey of Architectures, Enhancements, and Robustness Frontiers , author=. 2025 , eprint=

2025
[30]

2025 , eprint=

ComposeRAG: A Modular and Composable RAG for Corpus-Grounded Multi-Hop Question Answering , author=. 2025 , eprint=

2025
[31]

2025 , eprint=

TreeHop: Generate and Filter Next Query Embeddings Efficiently for Multi-hop Question Answering , author=. 2025 , eprint=

2025
[32]

2025 , eprint=

LevelRAG: Enhancing Retrieval-Augmented Generation with Multi-hop Logic Planning over Rewriting Augmented Searchers , author=. 2025 , eprint=

2025
[33]

2025 , eprint=

A Survey of Multimodal Retrieval-Augmented Generation , author=. 2025 , eprint=

2025
[34]

2025 , eprint=

Ask in Any Modality: A Comprehensive Survey on Multimodal Retrieval-Augmented Generation , author=. 2025 , eprint=

2025
[35]

2024 , eprint=

Hierarchical Retrieval-Augmented Generation Model with Rethink for Multi-hop Question Answering , author=. 2024 , eprint=

2024
[36]

Proceedings of the AAAI Conference on Artificial Intelligence , author=

A Multi-Agent Communication Framework for Question-Worthy Phrase Extraction and Question Generation , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2019 , month=. doi:10.1609/aaai.v33i01.33017168 , abstractNote=

work page doi:10.1609/aaai.v33i01.33017168 2019
[37]

Frontiers in Psychology , volume =

Reading bots: The implication of deep learning on guided reading , author =. Frontiers in Psychology , volume =. 2023 , doi =

2023
[38]

and Le, Quoc V

Wei, Jason and Wang, Xuezhi and Schuurmans, Dale and Bosma, Maarten and Ichter, Brian and Xia, Fei and Chi, Ed H. and Le, Quoc V. and Zhou, Denny , title =. Proceedings of the 36th International Conference on Neural Information Processing Systems , articleno =. 2022 , isbn =

2022
[39]

2025 , note =

Introduction to Text Embeddings at Cohere , howpublished =. 2025 , note =

2025
[40]

, title =

Lloyd, Stuart P. , title =. IEEE Transactions on Information Theory , volume =. 1982 , doi =

1982
[41]

2024 , eprint=

The Faiss library , author=. 2024 , eprint=

2024
[42]

Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

Reimers, Nils and Gurevych, Iryna. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. 2019

2019
[43]

Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages=

FanOutQA: A Multi-Hop, Multi-Document Question Answering Benchmark for Large Language Models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages=
[44]

Longbench v2: Towards deeper understanding and reasoning on realistic long-context multitasks, 2025

LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks , author=. arXiv preprint arXiv:2412.15204 , year=

work page arXiv
[45]

2024 , booktitle =

Leave No Document Behind: Benchmarking Long-Context LLMs with Extended Multi-Doc QA , author=. 2024 , booktitle =

2024
[46]

Gemini 2.5 Flash: Hybrid Reasoning with Controllable Thinking Budget , year =
[47]

1987 , issue_date =

Peter J. Rousseeuw , keywords =. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis , journal =. 1987 , issn =. doi:https://doi.org/10.1016/0377-0427(87)90125-7 , url =

work page doi:10.1016/0377-0427(87)90125-7 1987
[48]

and Bouldin, Donald W

Davies, David L. and Bouldin, Donald W. , journal=. A Cluster Separation Measure , year=
[49]

, author=

X-means: Extending k-means with efficient estimation of the number of clusters. , author=. Icml , volume=. 2000 , organization=

2000
[50]

International Conference on Neural Information Processing , pages=

G-means: a clustering algorithm for intrusion detection , author=. International Conference on Neural Information Processing , pages=. 2008 , organization=

2008
[51]

Pacific-Asia conference on knowledge discovery and data mining , pages=

Density-based clustering based on hierarchical density estimates , author=. Pacific-Asia conference on knowledge discovery and data mining , pages=. 2013 , organization=

2013
[52]

The Twelfth International Conference on Learning Representations , year=

Self-rag: Learning to retrieve, generate, and critique through self-reflection , author=. The Twelfth International Conference on Learning Representations , year=
[53]

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

Precise zero-shot dense retrieval without relevance labels , author=. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=
[54]

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

Query rewriting in retrieval-augmented large language models , author=. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

2023
[55]

arXiv preprint arXiv:2305.03653 , year=

Query expansion by prompting large language models , author=. arXiv preprint arXiv:2305.03653 , year=

work page arXiv
[56]

and Liu, Xuebo and Wong, Derek F

Chen, Guanhua and Yao, Yutong and Chao, Lidia S. and Liu, Xuebo and Wong, Derek F. SGIC : A Self-Guided Iterative Calibration Framework for RAG. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025. doi:10.18653/v1/2025.acl-long.1376

work page doi:10.18653/v1/2025.acl-long.1376 2025
[57]

Ammann, Paul J. L. and Golde, Jonas and Akbik, Alan. Question Decomposition for Retrieval-Augmented Generation. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop). 2025. doi:10.18653/v1/2025.acl-srw.32

work page doi:10.18653/v1/2025.acl-srw.32 2025