Pith. sign in

REVIEW 4 major objections 6 minor 41 references

Momentum Based Reward Design for Low Emission Traffic Signal Control

T0 review · 4 major / 6 minor · reviewed 2026-07-12 · grok-4.5

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

Pith's one-line read Progressive multi-turn text-to-visualization with rule-guided agent verification produces more accurate charts than one-shot translation.

desk verdict Solid progressive multi-turn NL2VIS systems paper with a real benchmark and clear gains; the soft spot is reverse-built trajectories, not the core empirical claim. read the letter →

arxiv 2605.29693 v2 pith:QFFDHUNF submitted 2026-05-28 cs.LG cs.RO

Chinmay Mundane , Amith Manoharan , Arun Kumar Singh This is my paper · ORCID
classification cs.LGcs.RO
keywords Text-to-VisualizationProgressivemulti-turninteractionMulti-agentcollaborationLargelanguagemodelsVisualizationqueryReActvalidationDataanalysis
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

Most text-to-visualization systems force users to specify chart type, filters, grouping, and aggregations in a single fully formed request. That one-shot setup raises cognitive load for non-experts and often yields charts that miss the real intent. This paper instead frames the task as progressive multi-turn refinement: every intermediate turn must produce a valid, executable visualization that later turns can extend. To study that setting, the authors build a trajectory benchmark by simplifying complete visualization queries under explicit structural and feasibility rules, then reconstructing natural-language steps. They further introduce a three-agent system whose validation agent repeatedly checks and repairs candidate queries with tools under syntax-first and schema-before-execution constraints. On the new benchmark, the full progressive-plus-validation design improves execution accuracy by up to about 18% in single-table and 23% in multi-table settings over strong one-shot baselines, while ablations show both multi-turn interaction and validation are needed for the gains.

What carries the argument

PMVisAgent: a User–System–Validation agent loop in which the Validation Agent runs a bounded ReAct-style tool cycle (syntax, schema, SQL execution, intent matching) under explicit non-leakage and precedence rules, together with formal reliability theorems that bound tool calls and guarantee safe execution.

What would settle it

Replace the scripted benchmark trajectories with free-form progressive sessions from new users who are not constrained to the reverse-derived question sequences, then re-measure whether the same progressive-plus-validation agent still beats matched one-shot baselines by similar execution-accuracy margins.

Watch

Extended reading notes

Core claim

The paper claims that progressive multi-turn text-to-vis—where each round emits a valid intermediate visualization query and later rounds add constraints—combined with a ReAct-style validation agent that repairs candidates under explicit interaction and tool-precedence rules, substantially outperforms traditional one-shot text-to-vis methods, reaching up to 88.6% and 77.86% execution accuracy in single-table and multi-table settings on the constructed progressive benchmark.

Load-bearing premise

The reverse-built multi-turn trajectories—made by randomly masking query clauses, regenerating natural language, and lightly correcting by hand—are assumed to be realistic enough proxies for how real users actually refine visualization requests.

Editorial extensions

If this is right

  • One-shot text-to-vis benchmarks cannot measure whether a system keeps intermediate charts valid while users add constraints over time.
  • Ordered tool use and non-leakage rules can give multi-agent visualization systems concrete reliability bounds on safety, tool-call count, and runtime.
  • Valid intermediate visualizations reduce error accumulation across dialogue rounds compared with waiting until a final complex query.
  • Existing one-shot visualization corpora can be extended into multi-turn benchmarks via constrained reverse simplification rather than only collecting new dialogues from scratch.
  • Progressive interaction and verification are complementary: removing either component measurably hurts end-to-end execution accuracy.

Reading between the lines

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

  • The same progressive-plus-verify pattern is a natural candidate for other structured generation tasks that suffer from early error compounding, such as multi-step text-to-SQL or iterative data-analysis agents.
  • If reverse-derived trajectories understate the messiness of real exploratory dialogue, live systems will need stronger online clarification policies than the benchmark alone reveals.
  • Keeping per-round prompts short while spreading work across turns may keep total cost competitive with heavy one-shot multi-agent pipelines.
  • Tool-precedence invariants (syntax before schema before execution) could become a reusable safety pattern whenever language models issue database side effects.
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

4 major / 6 minor

Summary. The paper proposes PMVis, a progressive multi-turn text-to-visualization paradigm in which each intermediate turn yields a valid, executable VQL that is refined as users add constraints. To support the paradigm, it constructs PMVisBench by reverse-simplifying VisEval terminal VQLs via randomized clause masking under Masking-Constraint and Visualization-Feasibility rules, then reconstructing NLQs with LLMs plus human correction. Building on this, it introduces PMVisAgent, a three-agent (User/System/Validation) framework with a ReAct-style tool loop and explicit UA/VA rules intended to limit information leakage and error accumulation. On PMVisBench, PMVisAgent reports large gains in execution accuracy over one-shot neural and LLM baselines (up to 88.60% single-table and 77.86% multi-table with qwen-plus), with ablations, cost analysis, a case study, and a small user study.

Significance. If progressive multi-turn refinement with intermediate executable visualizations is the right interaction model for non-expert analysis, this is a useful systems contribution: a trajectory-level benchmark, an agent design with explicit interaction/verification rules, and empirical evidence that multi-turn refinement plus validation improves execution correctness over strong one-shot baselines. Concrete strengths include the constrained reverse-construction pipeline (MC/VF rules), multi-backbone zero-shot evaluation, the three-way ablation isolating multi-turn vs validation (Table III), cost/latency comparison against nvAgent (Table V), and a user study that at least partially checks trajectory realism. The formal results (Theorems IV.1–IV.3) mainly restate rule-enforced tool precedence and bounded loops rather than end-task correctness; they are still valuable as engineering reliability bounds if presented as such.

major comments (4)
  1. [Section III-B; Section V-E; Table VI] Section III-B (Quality Assurance) and the user study (Section V-E, Table VI) are load-bearing for the central transfer claim. Intermediate trajectories are reverse-derived by randomized clause masking plus LLM NLQ reconstruction, not collected from live progressive analysis. The user study (20 students, 200 cases) is a helpful check (avg. 3.32 vs 3.7 rounds) but is too small and homogeneous to establish that reverse trajectories match real intent evolution, ordering of refinements, or language. Without stronger evidence (larger/more diverse user trajectories, or a live interactive evaluation), reported gains may overstate benefit in deployment. Please either expand the user study / collect forward trajectories, or substantially qualify claims about real-world progressive refinement.
  2. [Table II; Section V-B] Table II shows a severe gap between clause-level exact-match metrics and Execution Accuracy in multi-table settings (e.g., gpt-4o-mini: Overall ACC 5.13% vs Exec. ACC 65.35%; qwen-plus: 17.01% vs 77.86%). The text attributes this to “strict exact-match requirements,” but a ~5–17% overall match with ~65–78% execution match is not a minor syntactic discrepancy; it raises questions about what Exec. Acc. is actually measuring (result-table equality under alternative but valid VQLs? partial schema aliases? ordering/aggregation variants?). Please define execution equivalence precisely, quantify how often high Exec. Acc. coincides with wrong chart type/axes, and report failure modes. Without this, the multi-table headline gains are hard to interpret.
  3. [Section V-A.2; Table II] Baseline protocol fairness is under-specified for a multi-turn benchmark. One-shot models (Seq2Vis, ncNet, Prompt4Vis, etc.) appear evaluated on PMVisBench, but it is unclear whether they receive only the final NLQ, the concatenated history, or each intermediate NLQ independently, and whether intermediate executable targets are scored. If one-shot systems are forced into a setting they were not designed for without a strong multi-turn adaptation (e.g., history concatenation + re-generation), gains may partly reflect protocol mismatch rather than paradigm superiority. Please state the exact input each baseline sees at each turn and, if possible, add a strong multi-turn-adapted baseline beyond nvAgent/MMCoVisNet.
  4. [Section IV-D; Theorem IV.3] Theorems IV.1–IV.3 / Corollary IV.1.1 establish tool-action precedence, safe SQL execution under syntax+schema checks, and O(nm) tool-call/time bounds. These are useful control-flow invariants given VA-Rules 1–3, but they do not imply semantic correctness of the final VQL, non-accumulation of intent errors, or reliability of the overall text-to-vis mapping. The abstract/intro language about “theoretical reliability guarantees” and “System Reliability” (Theorem IV.3) can be read as stronger than the proofs deliver. Please reframe claims to match the actual theorems (safety of tool use and bounded cost), and avoid suggesting end-task correctness guarantees.
minor comments (6)
  1. [Table I] Table I lists Dial-NVBench as multi-turn but not progressive; the prose distinction is clear, but a short formal definition of “progressive” (executable intermediate VQL + monotonic addition of semantic components) in the table caption would help readers.
  2. [Section III-B] Figure 1 vs Figure 2 labeling in the text is slightly confusing (pipeline described as Fig. 2 in one place and Fig. 1 in another in the construction section). Please unify figure references.
  3. [Section IV-A, UA-Rule 2] UA-Rule 2 allows the User Agent to consult ground-truth VQL v*_i when answering clarifications, with a non-leak predicate. This is acceptable for simulation but should be stated more prominently as an evaluation assumption, since a real user would not have v*.
  4. [Section V-A.4; Table V] Implementation details fix m=10 ReAct steps and temperature 0; a brief sensitivity note on m (or early-stop rates) would strengthen the cost claims in Table V.
  5. [Sections III–IV] Minor writing issues: “I-th round” / notation consistency for H_i vs S_i; occasional grammar (“we obtain 1,149 distinct visualizations”); ensure all tool predicate projections (π_B, π_V) are defined before first use.
  6. [Abstract; Section V] Code link is provided; please also state whether PMVisBench trajectories and evaluation scripts for Exec. Acc. will be released in full to support reproduction of Table II’s execution metric.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical systems paper with no load-bearing circular derivation; theorems formalize rule-enforced control flow by construction, not claimed first-principles predictions.

  1. self definitional [Section IV-D, Theorems IV.1–IV.3 / VA-Rules 1–3]
    "Theorem IV.3 (System Reliability). For any n-round dialogue execution context Γ, PMVisAgent satisfies overall system reliability: Y(Γ)≜Θ_safety(Γ)∧Θ_tool(Γ)∧Θ_time(Γ), where Θ_safety(Γ) denotes execution safety, Θ_tool(Γ) denotes bounded tool-call complexity, and Θ_time(Γ) denotes bounded validation time."

    “Reliability” is defined as the conjunction of the three properties already enforced by VA-Rules (syntax-first, schema-before-execution, T≤m) and Algorithm 1’s single-action loop; the theorems then restate those constraints. This is definitional formalization of control flow, not a circular empirical prediction—and the paper presents it as such—so impact is minor and non-load-bearing for the accuracy claims.

full rationale

PMVis/PMVisAgent is an empirical multi-agent systems paper: the central claims are execution-accuracy gains on PMVisBench versus one-shot baselines (Tables II–III, V–VI), not a fitted-parameter or uniqueness-theorem derivation. Dataset construction (reverse VQL simplification under MC/VF rules, LLM NLQ reconstruction, human correction) co-designs a progressive benchmark with the progressive method, which is standard for introducing a new paradigm and is partially stress-tested by a small user study—not a reduction of the reported metric to its own inputs. UA-Rule 2 allows the simulated User Agent to consult ground-truth VQL only to produce non-leaking natural-language clarifications; that is evaluation scaffolding, not a self-definitional prediction of Exec Acc. Theorems IV.1–IV.3 prove tool-precedence, bounded tool calls (≤nm), and bounded validation time from VA-Rules 1–3 and Algorithm 1; they honestly define “reliability” as safety∧tool-bound∧time-bound and do not claim semantic correctness of generated VQLs. No fitted constants are renamed as predictions, no uniqueness theorem is imported from overlapping authors to force the result, and no known empirical pattern is merely renamed. Score 1 reflects only the mild, non-load-bearing co-design of paradigm and reverse-derived trajectories; the empirical claim remains independently supported by held-out execution match and ablations.

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

Load-bearing content is mostly engineering assumptions and evaluation design choices rather than free physical parameters. Reliability theorems depend on rule compliance and tool semantics; empirical claims depend on reverse-trajectory fidelity and LLM tool-loop behavior.

free parameters (3)
  • max ReAct steps m = 10
    Validation loop bound set to 10; affects repair budget, latency, and success rate.
  • LLM decoding settings = temperature=0; multiple commercial backbones
    temperature/frequency/presence penalties fixed to 0; backbone choice (gpt-4o-mini, gemini, qwen-plus) strongly affects reported accuracy.
  • random clause-masking trajectory sampler = uniform over optional clauses with rule rejection/resample
    Which optional clause is removed each step is stochastic under MC/VF acceptance; shapes the benchmark distribution of refinement paths.
assumptions (5)
  • domain assumption Real exploratory visualization is well modeled as progressive addition of VQL semantic components with a valid intermediate visualization each turn.
    Core PMVis formulation (Eqs. 2–4) and motivation in Introduction/Problem Formulation.
  • domain assumption Masking-Constraint and Visualization-Feasibility rules preserve structurally legal, renderable, non-empty intermediate VQLs that remain meaningful training/eval targets.
    Section III-B MC-Rule 1–3 and VF-Rule 1–2 define accepted simplifications.
  • ad hoc to paper LLM-generated then human-corrected NLQs aligned to reverse-simplified VQLs adequately simulate user language and refinement order.
    NLQ Construction and Quality Assurance; user study is only partial external validation.
  • domain assumption User-Agent clarification can use ground-truth VQL internally without leaking executable content, and Validation tools return faithful syntax/schema/exec/intent verdicts.
    UA-Rule 2–3 and tool definitions in Section IV; reliability theorems assume these semantics.
  • domain assumption Execution accuracy on assembled SQL/VQL results is a sufficient primary measure of visualization correctness for comparing systems.
    Metrics section; used as headline improvement despite weaker exact-match clause metrics in multi-table cases.
invented entities (3)
  • PMVis progressive multi-turn paradigm independent evidence
    purpose: Recast text-to-vis as a sequence of NLQ–VQL pairs with progressive refinement and intermediate executability.
    Central task reformulation distinguishing the work from one-shot NL2VIS.
  • PMVisBench
    purpose: Provide trajectory-level progressive sessions with intermediate valid VQLs derived from VisEval terminals.
    First dedicated progressive text-to-vis benchmark claimed by authors; evidence is internal construction plus small user study.
  • PMVisAgent (User/System/Validation agents with UA/VA rules)
    purpose: Simulate multi-turn dialogue and mitigate error accumulation via ReAct tool repair under explicit rules.
    Main method; performance evidence is empirical on PMVisBench, not independent external deployment.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Momentum Based Reward Design for Low Emission Traffic Signal Control." pith.science (2026). https://pith.science/paper/QFFDHUNF

@misc{pith2026260529693,
  author       = {Pith},
  title        = {Pith review of: Momentum Based Reward Design for Low Emission Traffic Signal Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QFFDHUNF}},
  note         = {Machine review of arXiv:2605.29693}
}
read the original abstract

Urban traffic congestion is a growing global issue contributing significantly to long commute times and environmental pollution. Traditional traffic signal control systems often fail to adapt to dynamic traffic conditions. Adaptive traffic signal control can improve urban traffic without changing road infrastructure. Deep Reinforcement Learning (DRL) has shown strong performance for this task, but existing delay and queue-based rewards often produce short-sighted or unstable policies. This paper proposes a Momentum-Based Reward Function (MBRF) that encourages vehicles to keep moving rather than penalizing congestion alone. The method is evaluated in SUMO (Simulation of Urban MObility) using standard traffic metrics such as waiting time, queue length, throughput, and CO2 emissions. Results show that the proposed reward produces better throughput-emission trade-offs and more stable learning behavior than delay or queue-based rewards, as well as classical controllers such as Max Pressure and LQF.

Figures

Figures reproduced from arXiv: 2605.29693 by the authors.

Figure 1
Figure 1. The traffic signal control problem is modeled as a Markov Decision [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Two-way single intersection with through, left, and right options [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 5
Figure 5. Throughput comparison across methods. 0 200 400 600 800 1000 Time step (seconds) 0 20 40 60 Average Travel Time (s) Average Travel Time Comparison Wait Queue Diff Wait Momentum MaxP LQF [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Average queue length during evaluation episodes. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 7
Figure 7. Figure 7: Total CO2 emissions comparison. B. Travel Time Performance Average travel time results further illustrate trade-offs be￾tween objectives. Although the waiting-time reward achieves the lowest travel time (22.6 s), it is accompanied by ex￾tremely poor queue and emission …

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 2 linked inside Pith

  1. [1]

    Synthesizing natural language to visualization (nl2vis) benchmarks from nl2sql benchmarks,

    Y . Luo, N. Tang, G. Li, C. Chai, W. Li, and X. Qin, “Synthesizing natural language to visualization (nl2vis) benchmarks from nl2sql benchmarks,” inProceedings of the 2021 ACM SIGMOD International Conference on Management of Data, 2021, pp. 1235–1247

    2021

  2. [2]

    Aurora: Data-driven construction of visual graph query interfaces for graph databases,

    S. S. Bhowmick, K. Huang, H. E. Chua, Z. Yuan, B. Choi, and S. Zhou, “Aurora: Data-driven construction of visual graph query interfaces for graph databases,” inProceedings of the 2020 ACM SIGMOD Interna- tional Conference on Management of Data, 2020, pp. 2689–2692

    2020

  3. [3]

    Multivis-agent: A multi- agent framework with logic rules for reliable and comprehensive cross- modal data visualization,

    J. Lu, Y . Song, C. Zhang, and R. C.-W. Wong, “Multivis-agent: A multi- agent framework with logic rules for reliable and comprehensive cross- modal data visualization,” vol. 4, no. 1. ACM, feb 2026

    2026

  4. [4]

    Haichart: Human and ai paired visualization system,

    Y . Xie, Y . Luo, G. Li, and N. Tang, “Haichart: Human and ai paired visualization system,”Proceedings of the VLDB Endowment, vol. 17, no. 11, pp. 3178–3191, 2024

    2024

  5. [5]

    Navigating data repositories: Utilizing line charts to discover relevant datasets,

    D. Ji, H. Luo, Z. Bao, and S. Culpepper, “Navigating data repositories: Utilizing line charts to discover relevant datasets,”Proceedings of the VLDB Endowment, vol. 17, no. 12, pp. 4289–4292, 2024

    2024

  6. [6]

    Seedb: Efficient data-driven visualization recommendations to support visual analytics,

    M. Vartak, S. Rahman, S. Madden, A. Parameswaran, and N. Polyzotis, “Seedb: Efficient data-driven visualization recommendations to support visual analytics,” inProceedings of the VLDB Endowment International Conference on Very Large Data Bases, vol. 8, no. 13, 2015, p. 2182

    2015

  7. [7]

    Crowdchart: Crowdsourced data extraction from visualization charts,

    C. Chai, G. Li, J. Fan, and Y . Luo, “Crowdchart: Crowdsourced data extraction from visualization charts,”IEEE Transactions on Knowledge and Data Engineering, vol. 33, no. 11, pp. 3537–3549, 2020

    2020

  8. [8]

    Steerable self- driving data visualization,

    Y . Luo, X. Qin, C. Chai, N. Tang, G. Li, and W. Li, “Steerable self- driving data visualization,”IEEE Transactions on Knowledge and Data Engineering, vol. 34, no. 1, pp. 475–490, 2020

    2020

Show all 41 references
  1. [9]

    Natural language interfaces for tabular data querying and visualization: A survey,

    W. Zhang, Y . Wang, Y . Song, V . J. Wei, Y . Tian, Y . Qi, J. H. Chan, R. C.- W. Wong, and H. Yang, “Natural language interfaces for tabular data querying and visualization: A survey,”IEEE Transactions on Knowledge and Data Engineering, vol. 36, no. 11, pp. 6699–6718, 2024

    2024

  2. [10]

    Visualization recommendation through visual relation learning and visual preference learning,

    D. Ji, H. Luo, and Z. Bao, “Visualization recommendation through visual relation learning and visual preference learning,” in2023 IEEE 39th International Conference on Data Engineering (ICDE). IEEE, 2023, pp. 1860–1873

    2023

  3. [11]

    Towards robustness of text-to-visualization translation against lexical and phrasal variability,

    J. Lu, Y . Song, H. Zhang, C. J. Zhang, K. Wu, and R. C.-W. Wong, “Towards robustness of text-to-visualization translation against lexical and phrasal variability,” in2025 IEEE 41st International Conference on Data Engineering (ICDE). IEEE Computer Society, 2025, pp. 793–806

    2025

  4. [12]

    Datavist5: A pre-trained language model for jointly understanding text and data visualization,

    Z. Wan, Y . Song, S. Li, C. J. Zhang, and R. C.-W. Wong, “Datavist5: A pre-trained language model for jointly understanding text and data visualization,” in2025 IEEE 41st International Conference on Data Engineering (ICDE). IEEE, 2025, pp. 1704–1717

    2025

  5. [13]

    Vega- lite: A grammar of interactive graphics,

    A. Satyanarayan, D. Moritz, K. Wongsuphasawat, and J. Heer, “Vega- lite: A grammar of interactive graphics,”IEEE transactions on visual- ization and computer graphics, vol. 23, no. 1, pp. 341–350, 2016

    2016

  6. [14]

    ggplot2: elegant graphics for data analysis,

    R. A. M. Villanueva and Z. J. Chen, “ggplot2: elegant graphics for data analysis,” 2019

    2019

  7. [15]

    Natural language to visualization by neural machine translation,

    Y . Luo, N. Tang, G. Li, J. Tang, C. Chai, and X. Qin, “Natural language to visualization by neural machine translation,”IEEE Transactions on Visualization and Computer Graphics, vol. 28, no. 1, pp. 217–226, 2021

    2021

  8. [16]

    nvbench 2.0: Resolving ambiguity in text-to-visualization through stepwise reasoning,

    T. Luo, C. Huang, L. Shen, B. Li, S. Shen, W. Zeng, N. Tang, and Y . Luo, “nvbench 2.0: Resolving ambiguity in text-to-visualization through stepwise reasoning,”arXiv preprint arXiv:2503.12880, 2025

    2025

  9. [17]

    Viseval: A benchmark for data visualization in the era of large language models,

    N. Chen, Y . Zhang, J. Xu, K. Ren, and Y . Yang, “Viseval: A benchmark for data visualization in the era of large language models,”IEEE Transactions on Visualization and Computer Graphics, 2024

    2024

  10. [18]

    Interactive text-to- visualization: Refining visualization outputs through natural language user feedback,

    X. Xiong, R. C.-W. Wong, and Y . Song, “Interactive text-to- visualization: Refining visualization outputs through natural language user feedback,” inProceedings of the 34th ACM International Confer- ence on Information and Knowledge Management, 2025, pp. 3571–3581

    2025

  11. [19]

    Marrying dialogue systems with data visualization: Interactive data visualization generation from natural language conversations,

    Y . Song, X. Zhao, and R. C.-W. Wong, “Marrying dialogue systems with data visualization: Interactive data visualization generation from natural language conversations,” inProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 2733– 2744

    2024

  12. [20]

    Rgvisnet: A hybrid retrieval-generation neural framework towards automatic data visualiza- tion generation,

    Y . Song, X. Zhao, R. C.-W. Wong, and D. Jiang, “Rgvisnet: A hybrid retrieval-generation neural framework towards automatic data visualiza- tion generation,” inProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022, pp. 1646–1655

    2022

  13. [21]

    Automated data visualization from natural language via large language models: An exploratory study,

    Y . Wu, Y . Wan, H. Zhang, Y . Sui, W. Wei, W. Zhao, G. Xu, and H. Jin, “Automated data visualization from natural language via large language models: An exploratory study,”Proceedings of the ACM on Management of Data, vol. 2, no. 3, pp. 1–28, 2024

    2024

  14. [22]

    nvAgent: Automated data visualization from natural language via collaborative agent workflow,

    G. Ouyang, J. Chen, Z. Nie, Y . Gui, Y . Wan, H. Zhang, and D. Chen, “nvAgent: Automated data visualization from natural language via collaborative agent workflow,” inProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers...

    2025

  15. [23]

    Automatic data visualization generation from chinese natural language questions,

    Y . Ge, V . J. Wei, Y . Song, J. C. Zhang, and R. C.-W. Wong, “Automatic data visualization generation from chinese natural language questions,” inProceedings of the 2024 Joint International Conference on Computa- tional Linguistics, Language Resources and Evaluation (LREC-COL...

    2024

  16. [24]

    Learning to recommend visualizations from data,

    X. Qian, R. A. Rossi, F. Du, S. Kim, E. Koh, S. Malik, T. Y . Lee, and J. Chan, “Learning to recommend visualizations from data,” inProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, 2021, pp. 1359–1369

    2021

  17. [25]

    Sevi: Speech-to- visualization through neural machine translation,

    J. Tang, Y . Luo, M. Ouzzani, G. Li, and H. Chen, “Sevi: Speech-to- visualization through neural machine translation,” inProceedings of the 2022 ACM SIGMOD International Conference on Management of Data, 2022, pp. 2353–2356

    2022

  18. [26]

    Datatone: Managing ambiguity in natural language interfaces for data visual- ization,

    T. Gao, M. Dontcheva, E. Adar, Z. Liu, and K. G. Karahalios, “Datatone: Managing ambiguity in natural language interfaces for data visual- ization,” inProceedings of the 28th annual acm symposium on user interface software & technology, 2015, pp. 489–500

    2015

  19. [27]

    Applying pragmatics principles for interaction with visual analytics,

    E. Hoque, V . Setlur, M. Tory, and I. Dykeman, “Applying pragmatics principles for interaction with visual analytics,”IEEE transactions on visualization and computer graphics, vol. 24, no. 1, pp. 309–318, 2017. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14

    2017

  20. [28]

    Flowsense: A natural language interface for visual data exploration within a dataflow system,

    B. Yu and C. T. Silva, “Flowsense: A natural language interface for visual data exploration within a dataflow system,”IEEE transactions on visualization and computer graphics, vol. 26, no. 1, pp. 1–11, 2019

    2019

  21. [29]

    Data2vis: Automatic generation of data visualizations using sequence-to-sequence recurrent neural networks,

    V . Dibia and C ¸ . Demiralp, “Data2vis: Automatic generation of data visualizations using sequence-to-sequence recurrent neural networks,” IEEE computer graphics and applications, vol. 39, no. 5, pp. 33–46, 2019

    2019

  22. [30]

    Chat2vis: Generating data visualizations via natural language using chatgpt, codex and gpt-3 large language models,

    P. Maddigan and T. Susnjak, “Chat2vis: Generating data visualizations via natural language using chatgpt, codex and gpt-3 large language models,”Ieee Access, vol. 11, pp. 45 181–45 193, 2023

    2023

  23. [31]

    prompt4vis: prompting large language models with example mining for tabular data visualization,

    S. Li, X. Chen, Y . Song, Y . Song, C. J. Zhang, F. Hao, and L. Chen, “prompt4vis: prompting large language models with example mining for tabular data visualization,”The VLDB Journal, vol. 34, no. 4, pp. 1–26, 2025

    2025

  24. [32]

    The design and implementation of xiaoice, an empathetic social chatbot,

    L. Zhou, J. Gao, D. Li, and H.-Y . Shum, “The design and implementation of xiaoice, an empathetic social chatbot,”Computational Linguistics, vol. 46, no. 1, pp. 53–93, 2020

    2020

  25. [33]

    Towards a human- like open-domain chatbot,

    D. Adiwardana, M.-T. Luong, D. R. So, J. Hall, N. Fiedel, R. Thoppilan, Z. Yang, A. Kulshreshtha, G. Nemade, Y . Luet al., “Towards a human- like open-domain chatbot,”arXiv preprint arXiv:2001.09977, 2020

    2001 arXiv

  26. [34]

    Blenderbot 3: a deployed con- versational agent that continually learns to responsibly engage,

    K. Shuster, J. Xu, M. Komeili, D. Ju, E. M. Smith, S. Roller, M. Ung, M. Chen, K. Arora, J. Laneet al., “Blenderbot 3: a deployed con- versational agent that continually learns to responsibly engage,”arXiv preprint arXiv:2208.03188, 2022

    2022 arXiv

  27. [35]

    Dialogue act recognition via crf-attentive structured network,

    Z. Chen, R. Yang, Z. Zhao, D. Cai, and X. He, “Dialogue act recognition via crf-attentive structured network,” inThe 41st international acm sigir conference on research & development in information retrieval, 2018, pp. 225–234

    2018

  28. [36]

    Recent advances and challenges in task-oriented dialog systems,

    Z. Zhang, R. Takanobu, Q. Zhu, M. Huang, and X. Zhu, “Recent advances and challenges in task-oriented dialog systems,”Science China Technological Sciences, vol. 63, no. 10, pp. 2011–2027, 2020

    2011

  29. [37]

    Recent neural methods on dialogue state tracking for task-oriented dialogue systems: A survey,

    V . Balaraman, S. Sheikhalishahi, and B. Magnini, “Recent neural methods on dialogue state tracking for task-oriented dialogue systems: A survey,” inProceedings of the 22nd annual meeting of the special interest group on discourse and dialogue, 2021, pp. 239–251

    2021

  30. [38]

    End-to-end task-completion neural dialogue systems,

    X. Li, Y .-N. Chen, L. Li, J. Gao, and A. Celikyilmaz, “End-to-end task-completion neural dialogue systems,” inProceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 2017, pp. 733–743

    2017

  31. [39]

    A network-based end-to-end trainable task-oriented dialogue system,

    T.-H. Wen, D. Vandyke, N. Mrk ˇsi´c, M. Gasic, L. M. R. Barahona, P.- H. Su, S. Ultes, and S. Young, “A network-based end-to-end trainable task-oriented dialogue system,” inProceedings of the 15th Conference of the European Chapter of the Association for Computational Linguist...

    2017

  32. [40]

    Ubar: Towards fully end-to-end task- oriented dialog system with gpt-2,

    Y . Yang, Y . Li, and X. Quan, “Ubar: Towards fully end-to-end task- oriented dialog system with gpt-2,” inProceedings of the AAAI confer- ence on artificial intelligence, vol. 35, no. 16, 2021, pp. 14 230–14 238

    2021

  33. [41]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in neural information processing systems, vol. 30, 2017

    2017

Pith tools

Reviewed July 12, 2026 · model on record in the stance chip above.