Pith. sign in

REVIEW 1 major objections 5 minor 2 cited by

Answer-Reconstruction Search Density: Measuring the Query and Source Work Compressed by Conversational Answers

T0 review · 1 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Conversational answers can be measured by the minimum number of distinct web queries and pages needed to reconstruct a fixed share of their content—a policy-relative proxy for the search work they replace.

desk verdict Clean set-cover formalization of answer compression, honestly labeled; the empirical magnitude is a lexical diagnostic, not measured search work — worth reviewing, but the headline should be read with the authors' own caveats. read the letter →

arxiv 2607.18904 v1 pith:7SVLOF4D submitted 2026-07-21 cs.IR

Benjamin Tannenbaum This is my paper
classification cs.IR
keywords answer-reconstructionsearchdensityconversationalsetcoverquerydecompositionretrievalevaluationmulti-turndialogueanswercompressioneffortmeasurement
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 that a conversational answer can be evaluated by a counterfactual: how many distinct conventional web queries and source pages would be needed, under a fixed rule, to reconstruct a fixed share of the answer's retrievable content. The central measure, answer-reconstruction search density, is the minimum size of such a query set, computed exactly as a partial set-cover problem. On 183 information-seeking conversations, the paper finds that answers are concentrated: a typical answer needs 3 lexical facets to cover 80% of its units. The paper also finds that multi-turn dialogues appear denser only because they are longer; after adjusting for answer size, dialogue depth itself adds little. The work matters because it turns the invisible search process hidden inside a generated answer into a countable quantity that can be audited, compared across systems, and separated into query work versus source work.

What carries the argument

The central object is the exact partial set-cover formulation (Eq. 1): given a set of candidate query actions and a binary support matrix indicating which answer units each query's results directly support, ARSD is the minimum number of queries whose union covers at least ⌈τm⌉ of the m units. The paper solves this exactly via bitmask dynamic programming over small candidate pools. Because live retrieval on the large corpus was blocked, the paper introduces a structural facet-cover diagnostic (FSD) that substitutes TF-IDF cosine similarity between answer units for the support matrix, treating each unit as a candidate prototype; FSD is explicitly not a query count and is reported separately. T

What would settle it

Run the approved private reconstruction protocol on the 183 conversations: issue real queries under the same policy and compare median live-web ARSD80 with the structural FSD80 median of 3. If they differ by more than one query, or if human-assigned facet boundaries disagree with TF-IDF boundaries on a majority of answers, the lexical-facet bridge is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that answer-reconstruction search density (ARSD) is a policy-relative measure of the minimum search work compressed into a synthesized answer. Formally, for a conversation, ARSD is the smallest number of distinct query actions needed so that the union of their returned evidence directly supports at least ⌈τm⌉ of the answer's m atomic retrievable units; a parallel page density does the same for distinct supporting pages. Empirically, in a consent-governed corpus of 183 information-seeking conversations, the median structural facet density is 3 (IQR 2–4) at an 80% coverage target under the primary lexical policy, with 3.25 units covered per selected facet. The mult

Load-bearing premise

The empirical backbone assumes that TF-IDF cosine similarity at a fixed threshold partitions answer units into the same facets that real web queries would need to cover; that link is untested, and the live reconstruction that could validate it was not executed.

Editorial extensions

If this is right

  • If ARSD is adopted, conversational answers become auditable in terms of the search operations they replace: an answer with density d can be said to compress at least d distinct query actions at a stated coverage and policy.
  • The finding that multi-turn depth does not independently raise density implies that dialogue length is a poor proxy for informational concentration; systems that extend conversations should be evaluated on answer volume rather than turn count alone.
  • The divergence between query density and page density means that answering systems can compress queries while still requiring many source pages, so evaluation should track both quantities rather than a single effort signal.
  • The rank-ordering stability across thresholds suggests ARSD can be used as a comparative measure even if the absolute scale shifts with the chosen policy, enabling cross-system comparison under a documented policy.
  • Aggregate ARSD over classes of answers could be used to translate observed conversational interactions into query-equivalent volumes, potentially reconciling keyword analytics with actual information demand.

Reading between the lines

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

  • If the FSD-to-ARSD bridge holds in a larger validation, the structural diagnostic becomes a cheap, retrieval-free way to estimate query compression across large answer corpora, since it needs only the answer text.
  • The measure could be extended to weight query actions by cost (e.g., result-click effort) instead of counting distinct queries, yielding a richer notion of compressed effort.
  • For RAG and agentic search systems, ARSD offers a testable objective: a system that generates a high-density answer should be able to justify it with a reconstruction trace, turning the metric into a provenance audit.
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

1 major / 5 minor

Summary. The paper defines answer-reconstruction search density (ARSD): the minimum number of distinct query actions, under a fixed and dated reconstruction policy, required to support a target share ⌈τm⌉ of an answer's atomic retrievable units, formalized as an exact partial set-cover problem (Eq. 1). A parallel page-density measure (PD) separates query compression from source compression. The authors evaluate the construct in two stages: (1) a structural facet-cover diagnostic (FSD, Eq. 3) over 183 consent-governed information-seeking conversations, reporting a median of 3 lexical facets at the primary policy, with sensitivity analyses across τ, γ, and unit caps; and (2) a public live-web calibration with 6 synthetic tasks and 36 queries, reporting median ARSD80 of 1.5 queries and median PD80 of 2 pages. The paper also examines whether multi-turn dialogue is associated with density after adjusting for answer volume, finding that the raw association attenuates substantially once unit count is controlled.

Significance. If the results hold, the paper contributes a policy-relative, answer-level measurement of compressed conventional search work, filling a gap between retrieval metrics that evaluate ranking/effort/support and the need to quantify the minimum query and source actions implied by a completed conversational answer. Strengths include the exact partial set-cover formulation; an exact bitmask solver that removes greedy-approximation error; a transparent two-dimensional sensitivity grid (τ 0.50–1.00 × γ 0.10–0.25), cap robustness, and bootstrap/permutation checks on the depth analysis; and fully public synthetic calibration materials. The authors are appropriately careful to label FSD as a lexical diagnostic rather than an observed query count, and they explicitly disclaim that the structural result should be reported as “three web searches.”

major comments (1)
  1. [§3.3, Eq. (3); §6.1; §7.1] The unit-extraction pipeline (§4.2) is deterministic and has no inter-rater audit, and the 16-unit cap affects 34.4% of conversations. Since FSD values, unit counts, and the depth adjustment all depend on this segmentation, plausible alternative segmentations could change the reported medians and attenuation results. The paper lists this as a limitation, but it is load-bearing for the empirical magnitude and for the depth conclusion; a small double-coding study and a segmentation-sensitivity analysis would materially strengthen the claims.
minor comments (5)
  1. [§5.1] The primary thresholds γ=0.15, τ=0.80, and K=16 are plausible but no rationale is given. The sensitivity grid mitigates this, but a sentence justifying the primary policy would help readers interpret the headline values.
  2. [§3.3, Eq. (3)] The TF-IDF document-frequency basis is stated in §5.1 (over 1,994 retained units), but specifying the exact tokenization and whether the vector is L2-normalized would improve reproducibility.
  3. [Figure 3 caption] The caption says “The bold cell is the primary policy,” but no cell is visibly bold in the text version. Please mark it explicitly or adjust the caption.
  4. [Table 4] McFadden pseudo-R² values are very small (0.015–0.068); the paper's interpretation is appropriately cautious, but the goodness-of-fit should be discussed in a sentence to avoid over-reading the model comparisons.
  5. [§2.2] The Huang et al. (2026) reference appears to be a self-citation or a very recent arXiv preprint; please verify that its year and venue are correct and that it is not an artifact of the arXiv ID.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ARSD and FSD are independently defined, no fitted parameter is relabeled as a prediction, and no self-citation chain is load-bearing.

full rationale

The paper's central construct, ARSD_τ(c;π), is defined as the optimum of an exact partial set-cover problem over a policy-dependent support matrix (Eq. 1). The structural diagnostic FSD is a separate definition (Eq. 3) that substitutes a TF-IDF cosine similarity matrix into the same optimization. The paper explicitly and repeatedly distinguishes FSD from live-web ARSD: FSD is 'not an observed query count and not a formal lower bound on live-web ARSD' (§3.3), and the live-web calibration is presented as feasibility evidence only, not as a corpus estimate (§5.3, §6.4, §8). The median-of-3 facet result is a descriptive statistic computed under a stated operationalization, not a prediction derived from a fitted parameter. RQ3's depth-attenuation finding is tested with OLS, Poisson, and within-stratum permutation analyses rather than assumed; no fitted coefficient is renamed as an independent prediction. There are no author self-citations carrying load-bearing weight, no imported uniqueness theorem, and no ansatz smuggled via citation. The strongest skeptical concern—that lexical FSD may not track true query-reconstruction work—is a construct-validity limitation that the paper acknowledges (third limitation, §8), and it does not constitute circularity under the rules because no step reduces, by definition or by self-citation, to its own inputs.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The reported magnitudes rest on four disclosed policy parameters (γ, τ, K, and extraction thresholds) whose grid moves the median density from 1 to 6, a single-coder extraction pipeline, a lexical proxy for facets, and an unreleased commercial corpus. None of these is hidden; rank-order robustness is shown, but the headline scale is policy-relative by design.

free parameters (5)
  • γ = 0.15 cosine similarity threshold (primary FSD policy) = 0.15 (sensitivity grid 0.10–0.25)
    Hand-chosen threshold defining lexical facet linkage in Eq. 3; at τ=0.80 it moves the median from 2 to 5 across the grid. Disclosed policy parameter, not fitted to a target.
  • τ = 0.80 target coverage = 0.80 (grid: 0.50, 0.80, 1.00)
    Coverage target defining ARSD80/FSD80; at γ=0.15 median rises 1→3→4 across the grid.
  • K = 16 answer-unit cap = 16 (robustness at 8 and 12)
    Sampling cap per conversation, bounding the set-cover state space at 2^16; mean density moves 3.09–3.28 and median stays 3. Affects 34.4% of cases, acknowledged.
  • Jaccard 0.78 unit-dedup threshold = 0.78
    Collapses near-duplicate units during extraction (§4.2); a choice that changes the effective unit set and therefore the cover solution.
  • Unit length window (4–48 non-stopword terms) = 4–48 terms
    Defines which sentences/list items become candidate answer units (§4.2); boundary choice in the deterministic pipeline.
assumptions (5)
  • domain assumption TF-IDF cosine similarity ≥ γ partitions an answer into query-relevant facets
    The whole FSD diagnostic (Eq. 3) assumes lexical similarity tracks the facets a real search would need. §3.3 concedes FSD is not a lower bound on live ARSD, and §8 says semantic validity is unestablished.
  • domain assumption Deterministic rule-based unit extraction yields atomic externally-retrievable units
    §4.2: sentence-splitting, list items, and Jaccard collapse, with no independent double-coding; Limitations state a sentence can contain multiple claims and a list item can be non-retrievable.
  • domain assumption The funnel (English-labeled, information-seeking, ≥3 retained units) defines the population of interest
    Table 1: 808 unique conversations reduce to 183 eligible cases; 351 records lack language labels. The paper scopes results to a 'selected corpus' of commercial relevance, not conversational AI generally.
  • domain assumption Binary support matrix M can be adjudicated reliably
    §5.3: a single reviewer assigned binary support for the public tasks; lexical overlap alone was insufficient. Inter-rater reliability was not measured, so M carries unquantified judgment error.
  • standard math Exact set cover via bitmask dynamic programming
    Appendix A: enumerating at most 2^16 support unions and keeping the minimum-cost cover per union is exact; NP-hardness of set cover is standard (Chvátal 1979). This is a computational fact, not an empirical claim.
invented entities (1)
  • ARSD / FSD / PD measurement constructs independent evidence
    purpose: New answer-level metrics of compressed conventional search work: the minimum number of query actions (ARSD), supporting pages (PD), or lexical prototypes (FSD) needed to cover τ of an answer's units under a fixed policy.
    Not physical entities. The constructs carry a falsifiable handle in the public live-web calibration (6 synthetic tasks, 36 fixed dated queries, audited URLs and support decisions) that any reviewer can rerun; the paper explicitly declines to claim corpus-level validity beyond that pilot.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Answer-Reconstruction Search Density: Measuring the Query and Source Work Compressed by Conversational Answers." pith.science (2026). https://pith.science/paper/7SVLOF4D

@misc{pith2026260718904,
  author       = {Pith},
  title        = {Pith review of: Answer-Reconstruction Search Density: Measuring the Query and Source Work Compressed by Conversational Answers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7SVLOF4D}},
  note         = {Machine review of arXiv:2607.18904}
}
read the original abstract

Conversational systems can collapse a visible sequence of web queries, result inspections, and source comparisons into a single synthesized answer. Existing retrieval metrics evaluate ranking, effort, or factual support, but they do not quantify the minimum conventional search work represented by a completed answer. We define answer-reconstruction search density (\ARSD): the minimum number of distinct query actions required, under a fixed and dated reconstruction policy, to support a target share of atomic retrievable answer units. A parallel page-density measure separates query compression from source compression.

Figures

Figures reproduced from arXiv: 2607.18904 by the authors.

Figure 1
Figure 1. The answer-first measurement perspective. The displayed values are corpus medians. Structural facets are not observed web queries; live query actions are measured separately. 1 Introduction A conventional web search exposes its process. The user refor￾mulates a query, opens results, compares sources, and decides when enough information has been collected. A conversational system can hide that sequence behind one res… view at source ↗
Figure 2
Figure 2. Full-population structural facet-density distribution (𝑁 = 183). The measure is an exact cover over lexical answer-unit prototypes, not live queries. are reported in facets; Poisson effects are exponentiated as incidence-rate ratios (IRRs). Percentile intervals use 5,000 OLS and 3,000 Poisson case-resampling draws. A functional-form￾free check estimates an overlap-weighted contrast within exact retained-unit-count s… view at source ↗
Figure 3
Figure 3. Robustness of structural facet density. Panel A varies target coverage and lexical similarity. The bold cell is the primary policy. Panel B re-estimates TF-IDF weights and exact covers under each unit cap. 3–5 6–9 10–12 13–15 16+ 2 4 Uncapped answer-unit bin Mean structural facet density Single turn Multi-turn (a) Mean density within answer-size bins; bars are bootstrap 95% intervals. 1 3 5 7 9 11 0 0.2 0.4 0.6 0.8 … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Depth, answer volume, and the density tail. Multi-turn cases dominate the long-answer bin, but within-bin means are similar from 6 units upward. 0.8 1.0 1.2 1.4 1.6 Full adjustment + answer units Unadjusted Multi-turn incidence-rate ratio [PITH_FULL_IMAGE:figures/full…
Figure 5
Figure 5. Figure 5: Poisson-model attenuation of the multi-turn association. Points are IRRs; bars are bootstrap 95% intervals. of those also need one page; the heat-pump task needs two pages. This demonstrates why query compression is not source compression. It also shows why the structu…
Figure 6
Figure 6. Figure 6: Query and page density in the public calibration. One broad query can expose evidence distributed across several source pages. or source can be absent from the opening prompt yet relevant to one reconstructed answer unit. Evaluation based on a single short keyword can …

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. The Prompt Is Not the Query: How Request State Evolves Across Multi-Turn AI Conversations

    cs.IR 2026-07 accept novelty 6.0 of 10

    In 8,133 real multi-turn AI conversations, the final user prompt contains only ~36% of the session's unique content vocabulary and reproduces the full detected request-state dimension set in only ~26% of dimension-bea...

  2. Beyond the Final Prompt: Measuring the Effect of Within-Conversation Context on AI Answers

    cs.IR 2026-08 conditional novelty 5.0 of 10

    With the final user message fixed, full-conversation and isolated-final answers differ materially in 44.7% of weighted cases; a compressed prefix shrinks the gap but leaves 30.8% materially different.

Reference graph

Works this paper leans on

35 extracted references · 2 canonical work pages · cited by 2 Pith papers

  1. [1]

    , title =

    Bates, Marcia J. , title =. Online Review , volume =. 1989 , doi =

    1989

  2. [2]

    Communications of the ACM , volume =

    Marchionini, Gary , title =. Communications of the ACM , volume =. 2006 , doi =

    2006

  3. [3]

    Psychological Review , volume =

    Pirolli, Peter and Card, Stuart , title =. Psychological Review , volume =. 1999 , doi =

    1999

  4. [4]

    , title =

    Broder, Andrei Z. , title =. SIGIR Forum , volume =. 2002 , doi =

    2002

  5. [5]

    and Kelly, Diane and Kim, Giyeong and Kim, Ja-Young and Lee, Hyuk-Jin and Muresan, Gheorghe and Tang, Muh-Chyun and Yuan, Xiaojun and Cool, Colleen , title =

    Belkin, Nicholas J. and Kelly, Diane and Kim, Giyeong and Kim, Ja-Young and Lee, Hyuk-Jin and Muresan, Gheorghe and Tang, Muh-Chyun and Yuan, Xiaojun and Cool, Colleen , title =. Proceedings of the 26th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2003 , doi =

    2003

  6. [6]

    Foundations and Trends in Information Retrieval , volume =

    Kelly, Diane , title =. Foundations and Trends in Information Retrieval , volume =. 2009 , doi =

    2009

  7. [7]

    , title =

    Jiang, Jiepu and Awadallah, Ahmed Hassan and Shi, Xiaolin and White, Ryen W. , title =. Proceedings of the Eighth ACM International Conference on Web Search and Data Mining , pages =. 2015 , doi =

    2015

  8. [8]

    and Awadallah, Ahmed Hassan and Dumais, Susan T

    Odijk, Daan and White, Ryen W. and Awadallah, Ahmed Hassan and Dumais, Susan T. , title =. Proceedings of the 24th ACM International Conference on Information and Knowledge Management , pages =. 2015 , doi =

    2015

Show all 35 references
  1. [9]

    ACM Transactions on Information Systems , volume =

    Moffat, Alistair and Bailey, Peter and Scholer, Falk and Thomas, Paul , title =. ACM Transactions on Information Systems , volume =. 2017 , doi =

    2017

  2. [10]

    Journal of the Association for Information Science and Technology , volume =

    Vuong, Tung and Saastamoinen, Miamaria and Jacucci, Giulio and Ruotsalo, Tuukka , title =. Journal of the Association for Information Science and Technology , volume =. 2019 , doi =

    2019

  3. [11]

    and Thomas, Paul and Mitra, Bhaskar and Sarkar, Shawon and Belkin, Nicholas , title =

    Shah, Chirag and White, Ryen W. and Thomas, Paul and Mitra, Bhaskar and Sarkar, Shawon and Belkin, Nicholas , title =. Proceedings of the 2023 Conference on Human Information Interaction and Retrieval , pages =. 2023 , doi =

    2023

  4. [12]

    and Dalton, Jeffrey and Radlinski, Filip , title =

    Zamani, Hamed and Trippas, Johanne R. and Dalton, Jeffrey and Radlinski, Filip , title =. Foundations and Trends in Information Retrieval , volume =. 2023 , doi =

    2023

  5. [13]

    Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Dalton, Jeffrey and Xiong, Chenyan and Kumar, Vaibhav and Callan, Jamie , title =. Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2020 , doi =

    2020

  6. [14]

    Findings of the Association for Computational Linguistics: EMNLP 2020 , pages =

    Kumar, Vaibhav and Callan, Jamie , title =. Findings of the Association for Computational Linguistics: EMNLP 2020 , pages =. 2020 , doi =

    2020

  7. [15]

    and Al Lawati, Sara Fahad Dawood and Mackenzie, Joel and Gallagher, Luke , title =

    Trippas, Johanne R. and Al Lawati, Sara Fahad Dawood and Mackenzie, Joel and Gallagher, Luke , title =. Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2024 , doi =

    2024

  8. [16]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics , pages =

    Trivedi, Harsh and Balasubramanian, Niranjan and Khot, Tushar and Sabharwal, Ashish , title =. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics , pages =. 2023 , doi =

    2023

  9. [17]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , year =

    Shao, Zhihong and Gong, Yeyun and Shen, Yelong and Huang, Minlie and Duan, Nan and Chen, Weizhu , title =. Findings of the Association for Computational Linguistics: EMNLP 2023 , year =

    2023

  10. [18]

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

    Min, Sewon and Krishna, Kalpesh and Lyu, Xinxi and Lewis, Mike and Yih, Wen-tau and Koh, Pang and Iyyer, Mohit and Zettlemoyer, Luke and Hajishirzi, Hannaneh , title =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages =. 2023 , doi =

    2023

  11. [19]

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

    Gao, Tianyu and Yen, Howard and Yu, Jiatong and Chen, Danqi , title =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages =. 2023 , doi =

    2023

  12. [20]

    Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics , pages =

    Abbasiantaeb, Zahra and Lupart, Simon and Aliannejadi, Mohammad , title =. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics , pages =. 2026 , doi =

    2026

  13. [21]

    Findings of the Association for Computational Linguistics: EMNLP 2024 , year =

    Song, Yixiao and Kim, Yekyung and Iyyer, Mohit , title =. Findings of the Association for Computational Linguistics: EMNLP 2024 , year =

    2024

  14. [22]

    arXiv preprint arXiv:2503.07919 , year =

    Song, Yixiao and Thai, Katherine and Pham, Chau Minh and Chang, Yapei and Nadaf, Mazin and Iyyer, Mohit , title =. arXiv preprint arXiv:2503.07919 , year =

    arXiv

  15. [23]

    Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages =

    Honovich, Or and Choshen, Leshem and Aharoni, Roee and Neeman, Ella and Szpektor, Idan and Abend, Omri , title =. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages =. 2021 , publisher =. doi:10.18653/v1/2021.emnlp-main.619 , url =

    2021 doi

  16. [24]

    and Choshen, Leshem and Astudillo, Ramon and Broderick, Tamara and Andreas, Jacob , title =

    Huang, Jenny Y. and Choshen, Leshem and Astudillo, Ramon and Broderick, Tamara and Andreas, Jacob , title =. arXiv preprint arXiv:2602.24287 , year =. doi:10.48550/arXiv.2602.24287 , url =

    doi

  17. [25]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) , pages =

    Don-Yehiya, Shachar and Choshen, Leshem and Abend, Omri , title =. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) , pages =. 2025 , publisher =. doi:10.18653/v1/2025.acl-demo.17 , url =

    2025 doi

  18. [26]

    Proceedings of the 2017 Conference on Conference Human Information Interaction and Retrieval , pages =

    Radlinski, Filip and Craswell, Nick , title =. Proceedings of the 2017 Conference on Conference Human Information Interaction and Retrieval , pages =. 2017 , doi =

    2017

  19. [27]

    Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics , pages =

    Petcu, Roxana and Murray, Kenton and Khashabi, Daniel and Kanoulas, Evangelos and de Rijke, Maarten and Lawrie, Dawn and Duh, Kevin , title =. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics , pages =. 2026 , publishe...

    2026 doi

  20. [28]

    arXiv preprint arXiv:2112.09332 , year =

    Nakano, Reiichiro and Hilton, Jacob and Balaji, Suchir and Wu, Jeff and Ouyang, Long and Kim, Christina and Hesse, Christopher and Jain, Shantanu and Kosaraju, Vineet and Saunders, William and Jiang, Xu and Cobbe, Karl and Eloundou, Tyna and Krueger, Gretchen and Button, Kevin...

    doi arXiv

  21. [29]

    , title =

    Wei, Jerry and Yang, Chengrun and Song, Xinying and Lu, Yifeng and Hu, Nathan and Huang, Jie and Tran, Dustin and Peng, Daiyi and Liu, Ruibo and Huang, Da and Du, Cosmo and Le, Quoc V. , title =. arXiv preprint arXiv:2403.18802 , year =. doi:10.48550/arXiv.2403.18802 , url =

    doi arXiv

  22. [30]

    Clarke, Charles L. A. and Kolla, Maheedhar and Cormack, Gordon V. and Vechtomova, Olga and Ashkan, Azin and B. Novelty and Diversity in Information Retrieval Evaluation , booktitle =. 2008 , doi =

    2008

  23. [31]

    A Greedy Heuristic for the Set-Covering Problem , journal =

    Chv. A Greedy Heuristic for the Set-Covering Problem , journal =. 1979 , doi =

    1979

  24. [32]

    and Stohr, Edward A

    Jarke, Matthias and Turner, Jon A. and Stohr, Edward A. and Vassiliou, Yannis and White, Norman and Michielsen, Kathleen , title =. IEEE Transactions on Software Engineering , volume =. 1985 , doi =

    1985

  25. [33]

    Long Range Planning , volume =

    Mazzelli, Ambra and De Massis, Alfredo and Petruzzelli, Antonio Messeni and Del Giudice, Manlio and Khan, Zaheer , title =. Long Range Planning , volume =. 2020 , doi =

    2020

  26. [34]

    , title =

    Kalbaugh, George E. , title =. Operations Research , volume =. 1993 , doi =

    1993

  27. [35]

    Nature Human Behaviour , volume =

    Salvi, Francesco and Horta Ribeiro, Manoel and Gallotti, Riccardo and West, Robert , title =. Nature Human Behaviour , volume =. 2025 , doi =

    2025

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.