pith. sign in

arxiv: 2605.27071 · v1 · pith:4PDCLZFNnew · submitted 2026-05-26 · 💻 cs.AI

Traceable Knowledge Graph Reasoning Enables LLM-Assisted Decision Support for Industrial VOCs in the Steel Industry

Changqing Su , Yu Ding , Zuhong Lin , Hongyu Liu , Xi He , Zheng Zeng , Liqing Li This is my paper

Pith reviewed 2026-06-29 17:37 UTC · model grok-4.3

classification 💻 cs.AI
keywords knowledge graphlarge language modelsVOCssteel industrydecision supportmulti-agent systemenvironmental informaticstraceable reasoning
0
0 comments X

The pith

A traceable knowledge graph from steel-industry VOCs literature powers a multi-agent LLM system with 96.93 percent precision on specialized questions.

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

The paper builds Chat-ISV to convert scattered scientific papers on volatile organic compounds in steel production into a structured Neo4j graph that links processes, pollutants, and control technologies. It adds topology optimization, source-backtracking retrieval, and multi-agent routing so the LLM can answer low-frequency industrial queries while citing its sources. The system is tested on 400 expert blind evaluations and shows large gains in factual reliability over plain LLMs. A sympathetic reader cares because the work demonstrates a concrete method for making large language models usable for real environmental-engineering decisions in narrow industrial domains.

Core claim

Chat-ISV parses a curated steel-industry VOCs literature corpus, builds a Neo4j knowledge graph containing 27180 nodes and 81779 semantic edges, applies chunk-centered topology optimization to cut isolated nodes from 57 percent to 4.08 percent, and combines this graph with prompt-constrained extraction, multi-agent routing, source-backtracking, local literature retrieval, and subgraph visualization to reach 96.93 percent precision, 72.63 percent recall, and an F1-score of 0.830 with a mean expert score of 1.69 out of 2.00.

What carries the argument

The Neo4j knowledge graph with chunk-centered topology optimization that supplies traceable, queryable links among processes, pollutants, and control technologies to constrain LLM outputs.

If this is right

  • Fragmented environmental-engineering literature becomes queryable and decision-support-oriented knowledge.
  • Source-backtracking retrieval lowers hallucination risk on low-frequency industrial questions.
  • Interactive subgraph visualization supports pollution-control decision making.
  • The same pipeline offers a scalable environmental-informatics paradigm for reliable LLM use in other specialized domains.

Where Pith is reading between the lines

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

  • The same graph-construction plus topology-optimization steps could be applied to VOCs or emissions data in other heavy industries such as petrochemicals or cement.
  • Adding real-time sensor or regulatory-update feeds into the graph might turn the system into a live monitoring aid rather than a static literature tool.
  • If the precision holds on out-of-corpus queries, the approach could serve as a template for regulated industries that must justify AI-assisted decisions with traceable evidence.

Load-bearing premise

The curated steel-industry VOCs literature corpus is representative and complete enough that the resulting knowledge graph captures the necessary process, pollutant, and control-technology relationships without critical omissions or extraction errors.

What would settle it

Running the system on a fresh set of steel-industry VOCs papers published after the corpus was built and checking whether the answers remain accurate or begin to hallucinate facts that the new papers contradict.

Figures

Figures reproduced from arXiv: 2605.27071 by Changqing Su, Hongyu Liu, Liqing Li, Xi He, Yu Ding, Zheng Zeng, Zuhong Lin.

Figure 1
Figure 1. Figure 1: Pipeline of the Chat-ISV system. Unstructured knowledge from domain￾specific scientific literature is extracted to construct a structured KG, and graph-retrieved evidence is incorporated into the LLM for question answering on VOCs governance in the steel industry. 2 Methodology 2.1 Data Acquisition and Information Extraction We used the Web of Science (WoS) database as the primary source to systematically … view at source ↗
Figure 2
Figure 2. Figure 2: b aligns synonymous entities, resolves semantic ambiguity, and reconstructs directional relations among sources, pollutants, and control routes. This semantic structure provides a more coherent overview of the field and may support cautious hypothesis generation. (a) (b) (c) (d) [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Local comparison of graph topology optimization. 3.3 Topology-Induced Reasoning in the Steel-Industry VOCs Knowledge Space Building on the topology-enhanced traceability established above, we next examined the reasoning role of the optimized KG topology. Specifically, we tested whether this topology could induce stepwise reasoning in the highly vertical steel-industry VOCs domain through constrained graph-… view at source ↗
Figure 4
Figure 4. Figure 4: a, we provide a zero-code, user-friendly web interface at https://huggingface.co/spaces/Grace-Ding/VOCs-graph. Through this interface, domain scholars can interactively query the VOCs-governance knowledge base and explore the Chat-ISV service without coding expertise. (b) (c) Based on the knowledge graph records, the characteristic VOCs emitted by the sintering process in the iron and steel industry mainly… view at source ↗
Figure 5
Figure 5. Figure 5: Comprehensive cross-model performance evaluation. 11/36 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: c showed pairwise Pearson correlations of 0.57–0.77 among the four experts, with the highest consistency between Experts B and C (0.77; Figure S3). These moderate-to-high correlations reflect broadly convergent expert judgment. Together, the high precision, low FP rate, and stable inter-expert consistency support the factual reliability of Chat-ISV for complex vertical-domain Q & A tasks. The extended-corp… view at source ↗
Figure 7
Figure 7. Figure 7: Chunk-centered star-topology local schematic in the Chat-ISV system. 22/36 [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Schema visualization and database statistics of the optimized knowledge graph in Neo4j. The completed topology integrates 27180 entity nodes and 81779 semantic relationship edges. 23/36 [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Snapshot of the initial knowledge-graph topology generated by the baseline LLM extraction pipeline. The visualization shows severe fragmentation with an isolated node rate of approximately 57 percent. 24/36 [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comprehensive quantitative evaluation and scoring consistency analysis based on expert blind assessment. Distribution of expert evaluation scores detailing the count of items assigned 0, 1, and 2 points by each of the four experts (a); Expert scoring consensus analysis presenting the Pearson correlation matrix across the four experts (b). 25/36 [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Generalization performance evaluation of the multi-agent system across broad environmental engineering topics and cross domain industrial scenarios. Heatmap of Precision, Recall, and F1-score evaluated independently by four experts (a); Count distribution matrix of true positives, false negatives, and false positives (b); Panoramic expert consensus matrix recording individual scoring variations across 100… view at source ↗
read the original abstract

Key knowledge for steel-industry volatile organic compounds (VOCs) governance is scattered across unstructured scientific literature, making it difficult to integrate process, pollutant, and control-technology evidence and increasing the risk of hallucination when general large language models (LLMs) answer low-frequency industrial questions. Here we developed Chat-ISV, a knowledge graph (KG) enhanced multi-agent Q&A system that parses a curated steel-industry VOCs literature corpus, constructs a Neo4j KG with 27180 nodes and 81779 semantic edges, and combines prompt-constrained extraction, chunk-centered topology optimization, multi-agent routing, source-backtracking retrieval, local literature retrieval, open-domain knowledge access, and interactive subgraph visualization. Benchmark tests and 400 expert blind evaluations showed that topology optimization reduced isolated nodes from 57% to 4.08% and that Chat-ISV achieved high factual reliability, with 96.93% precision, 72.63% recall, an F1-score of 0.830, and a mean score of 1.69/2.00. By converting fragmented environmental-engineering literature into traceable, queryable, and decision-support-oriented knowledge, Chat-ISV establishes a scalable environmental-informatics paradigm for reliable LLM deployment and intelligent pollution-control decision support in specialized industrial domains.

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.

Referee Report

3 major / 2 minor

Summary. The paper introduces Chat-ISV, a knowledge-graph-enhanced multi-agent Q&A system for steel-industry VOCs governance. It parses a curated literature corpus to build a Neo4j KG (27,180 nodes, 81,779 edges), applies chunk-centered topology optimization, multi-agent routing, source-backtracking retrieval, and interactive visualization. Benchmark tests and 400 expert blind evaluations are reported to show topology optimization reducing isolated nodes from 57% to 4.08%, with Chat-ISV achieving 96.93% precision, 72.63% recall, F1=0.830, and mean expert score 1.69/2.00, positioning the system as a scalable paradigm for traceable, hallucination-resistant decision support in specialized industrial domains.

Significance. If the corpus completeness and evaluation rigor hold, the work demonstrates a concrete, traceable KG+LLM pipeline that converts fragmented environmental-engineering literature into queryable decision-support knowledge, with measurable gains in factual reliability and reduced isolation in the graph. The combination of topology optimization, multi-agent routing, and expert blind validation provides a reproducible template for domain-specific LLM deployment where general models risk hallucination on low-frequency industrial questions.

major comments (3)
  1. [Abstract and Methods (corpus construction)] The central performance claims (96.93% precision, 72.63% recall, F1=0.830, expert mean 1.69/2) rest on an unverified assumption that the curated steel-industry VOCs literature corpus is representative and complete. No protocol is supplied for source selection, deduplication, coverage audit against known review papers, or post-extraction error sampling; if recent control-technology studies or regional literature are systematically omitted, both the topology-optimization result and the Q&A metrics become conditional on an untested completeness assumption.
  2. [Abstract and Evaluation section] Evaluation protocol details are absent: the abstract states specific performance numbers from 400 expert blind evaluations and topology optimization, but provides no information on how recall was measured against a gold standard, inter-rater agreement, question sampling strategy, or blinding procedure. This directly undermines the factual-reliability claims.
  3. [KG construction and topology optimization] The claim that the KG 'faithfully encodes the domain' for decision support is load-bearing, yet the manuscript supplies no quantitative validation (e.g., coverage sampling against external review papers or expert audit of extracted relations) that would confirm absence of critical gaps in process-pollutant-control relationships.
minor comments (2)
  1. [Results] Clarify the exact definition and measurement of 'isolated nodes' before and after topology optimization; the 57% to 4.08% reduction is a key result but the metric is not defined in the provided text.
  2. [System architecture] The multi-agent routing and source-backtracking mechanisms are described at a high level; a diagram or pseudocode would improve reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below and will revise the manuscript accordingly to provide the requested protocols and validations.

read point-by-point responses

  1. Referee: [Abstract and Methods (corpus construction)] The central performance claims (96.93% precision, 72.63% recall, F1=0.830, expert mean 1.69/2) rest on an unverified assumption that the curated steel-industry VOCs literature corpus is representative and complete. No protocol is supplied for source selection, deduplication, coverage audit against known review papers, or post-extraction error sampling; if recent control-technology studies or regional literature are systematically omitted, both the topology-optimization result and the Q&A metrics become conditional on an untested completeness assumption.

    Authors: We agree that the manuscript does not supply a full protocol for corpus construction. The original text refers only to a 'curated' corpus without detailing selection criteria or audits. In the revised manuscript we will add a dedicated Methods subsection describing the literature search strategy (databases, keywords, date range), inclusion/exclusion criteria, deduplication procedure, and any coverage or error-sampling steps performed. This addition will allow readers to evaluate the representativeness assumption directly. revision: yes

  2. Referee: [Abstract and Evaluation section] Evaluation protocol details are absent: the abstract states specific performance numbers from 400 expert blind evaluations and topology optimization, but provides no information on how recall was measured against a gold standard, inter-rater agreement, question sampling strategy, or blinding procedure. This directly undermines the factual-reliability claims.

    Authors: The evaluation details were condensed in the initial submission. We will expand the Evaluation section to specify the question sampling strategy, the construction of the gold-standard answers used for recall, the blinding procedure applied to the 400 expert evaluations, and inter-rater agreement metrics. These additions will make the reported precision, recall, and expert scores fully reproducible. revision: yes

  3. Referee: [KG construction and topology optimization] The claim that the KG 'faithfully encodes the domain' for decision support is load-bearing, yet the manuscript supplies no quantitative validation (e.g., coverage sampling against external review papers or expert audit of extracted relations) that would confirm absence of critical gaps in process-pollutant-control relationships.

    Authors: We acknowledge that the manuscript provides no external quantitative validation of KG coverage beyond the internal topology-optimization metrics. In the revision we will add a validation subsection that reports coverage sampling against external review papers and/or expert audit of a random subset of extracted relations, confirming the absence or presence of gaps in process-pollutant-control relationships. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on external expert validation

full rationale

The manuscript presents an applied KG+LLM system whose headline metrics (precision 96.93%, recall 72.63%, expert mean 1.69/2.00 from 400 blind evaluations) are obtained from independent human raters and node-isolation statistics. No equations, fitted parameters, or self-referential definitions appear in the abstract or described derivation. The single load-bearing assumption (corpus representativeness) is an empirical completeness claim, not a definitional or self-citation loop that forces the reported performance numbers. Therefore the result does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper introduces no free parameters, new physical entities, or ad-hoc axioms beyond the standard assumption that a curated domain corpus can be turned into a faithful semantic graph.

axioms (1)
  • domain assumption The curated steel-industry VOCs literature corpus is representative and complete enough that the resulting knowledge graph captures the necessary process, pollutant, and control-technology relationships without critical omissions or extraction errors.
    The entire pipeline depends on this corpus being sufficient; the abstract states the corpus was curated but gives no independent verification of coverage.

pith-pipeline@v0.9.1-grok · 5779 in / 1290 out tokens · 40590 ms · 2026-06-29T17:37:53.957842+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 1 canonical work pages · 1 internal anchor

  1. [1]

    Agrawal et al

    A. Agrawal et al. Large language models for knowledge graph extraction from scientific literature.Patterns, 5(1):100913, 2024

    2024

  2. [2]

    X. Bai, S. He, Y. Li, Y. Xie, X. Zhang, W. Du, and J.-R. Li. Construction of a knowledge graph for framework material enabled by large language models and its application.npj Computational Materials, 11:51, 2025

    2025

  3. [3]

    A. M. Bran et al. Chemcrow: Augmenting large-language models with chemistry tools.Nature Machine Intelligence, 6(5):481–487, 2024

    2024

  4. [4]

    Chen et al

    Z. Chen et al. Multi-pollutant collaborative control technologies in the iron and steel sintering process: Status and perspectives.Sci. Total Environ., 908:168291, 2024

    2024

  5. [5]

    Y. Ding. Dataset for Chat-ISV: A knowledge graph-enhanced large language model for question answering on VOC emission control in the steel industry. Zenodo, 2026

    2026

  6. [6]

    Dreger, K

    M. Dreger, K. Malek, and M. Eikerling. Large language models for knowledge graph extraction from tables in materials science.Digital Discovery, 4:1221, 2025

    2025

  7. [7]

    Y. Duan, Y. Tian, S. Ghosh, V. Venugopal, J. Chen, and E. A. Olivetti. Llm-empowered literature mining for material substitution studies in sustainable concrete.RESOURCES CONSER V ATION AND RECYCLING, 221:108379, 2025

    2025

  8. [8]

    Edge et al

    D. Edge et al. From local to global: A graph RAG approach to query-focused summarization.arXiv, 2024

    2024

  9. [9]

    Gao et al

    Y. Gao et al. Retrieval-augmented generation for large language models: A survey.arXiv, 2023

    2023

  10. [10]

    X. He, Y. Tian, Y. Sun, N. V. Chawla, T. Laurent, Y. LeCun, X. Bresson, and B. Hooi. G-retriever: Retrieval-augmented generation for textual graph understanding and question answering.Advances in Neural Information Processing Systems, 37, 2024

    2024

  11. [11]

    Hogan et al

    A. Hogan et al. Knowledge graphs.ACM Computing Surveys, 54(4):1–37, 2021

    2021

  12. [12]

    Huang et al

    L. Huang et al. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions.arXiv, 2023

    2023

  13. [13]

    Jafarzadeh, F

    P. Jafarzadeh, F. Ensan, M. A. A. Alavi, and F. Zarrinkalam. A knowledge graph embedding model for answering factoid entity questions.ACM Transactions on Information Systems, 43(2):Article 34, 2024

    2024

  14. [14]

    S. Ji, S. Pan, E. Cambria, P. Marttinen, and P. S. Yu. A survey on knowledge graphs: Representation, acquisition, and applications.IEEE Transactions on Neural Networks and Learning Systems, 33(2):494–514, 2022

    2022

  15. [15]

    Ji et al

    Z. Ji et al. Survey of hallucination in natural language generation.ACM Computing Surveys, 55(12):1–38, 2023

    2023

  16. [16]

    Kumar, F

    A. Kumar, F. Nazemi, H. Kodamana, M. Ramteke, and B. R. Bakshi. A large language model-based framework to retrieve life cycle inventory and environmental impact data from scientific literature.ENVIRONMENTAL SCIENCE & TECHNOLOGY, 59(42):22533–22543, 2025. 33/36

    2025

  17. [17]

    Y. Lan, Z. Hu, L. Wang, Y. Wang, D. Ye, P. Zhao, E.-P. Lim, H. Xiong, and H. Wang. LLM-based agent society investigation: Collaboration and confrontation in Avalon gameplay. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 100–113, 2024

    2024

  18. [18]

    Lewis et al

    P. Lewis et al. Retrieval-augmented generation for knowledge-intensive NLP tasks.Advances in Neural Information Processing Systems, 33:9459–9474, 2020

    2020

  19. [19]

    S. Li, J. Liao, Z. Zhang, and G. Xu. Catalytic decomposition of methane over Fe2O3-Al2O3 catalysts with high iron contents and at high CH4 space velocities. Resources Chemicals and Materials, 4(4):100123, 2025

    2025

  20. [20]

    Y. Li, D. Song, Y. Tian, H. Wang, C. Zhou, and S. Zhang. A framework of knowledge graph-enhanced large language model based on global planning.IEEE Transactions on Knowledge and Data Engineering, 38(2):736–748, 2026

    2026

  21. [21]

    M. Lin, T. Wu, and K. Xie. Low-carbon scheduling for power-hydrogen integrated energy system using large language model enhanced deep reinforcement learning.IEEE Transactions on Sustainable Energy, 2026

    2026

  22. [22]

    Z. Lin, J. Chen, Y. Fang, S.-h. Deng, H. Li, Y. Yang, and J. Yao. Rapidly tailor metal–organic frameworks for arsenate removal using graph convolutional neural networks.Separation and Purification Technology, 354:129334, 2025

    2025

  23. [23]

    Z. Lin, D. Ren, K. Ran, J. Sun, S. Yu, X. Bai, X. Huang, H. He, P. Pan, Y. Fang, Z. Li, H. Li, and J. Yao. Reshaping MOFs synthesis conditions mining with a dynamic multi-agents framework of large language model.Transactions of Materials Research, 2(1):100176, 2026

    2026

  24. [24]

    Liu et al

    Z. Liu et al. Characterizing the emission behaviors of cumulative VOCs from automotive solvent-based paint sludge.Journal of Environmental Management, 317:115369, 2022

    2022

  25. [25]

    Miret et al

    S. Miret et al. From text to insight: Large language models for chemical data extraction.Chemical Society Reviews, 54(3):1125–1150, 2025

    2025

  26. [26]

    Ouyang et al

    L. Ouyang et al. Training language models to follow instructions with human feedback.Advances in Neural Information Processing Systems, 35:27730–27744, 2022

    2022

  27. [27]

    S. Pan, L. Luo, Y. Wang, C. Chen, J. Wang, and X. Wu. Unifying large language models and knowledge graphs: A roadmap.IEEE Transactions on Knowledge and Data Engineering, 36(8):3534–3554, 2024

    2024

  28. [28]

    C. Su, Y. Guo, H. Chen, J. Zou, Z. Zeng, and L. Li. VOCs adsorption of resin-based activated carbon and bamboo char: porous characterization and nitrogen-doped effect.Colloids and Surfaces A: Physicochemical and Engineering Aspects, 601:124983, 2020

    2020

  29. [29]

    C. Su, Y. Guo, L. Yu, J. Zou, Z. Zeng, and L. Li. Insight into specific surface area, microporosity and N, P co-doping of porous carbon materials in the acetone adsorption.Materials Chemistry and Physics, 258:123930, 2020

    2020

  30. [30]

    C. Su, W. Jiang, Y. Guo, G. Yi, Z. Li, and H. Li. Rational molecular design of P-doped porous carbon material for the VOCs adsorption.Chinese Journal of Chemical Engineering, 79:155–163, 2025. 34/36

    2025

  31. [31]

    J. Sun, C. Xu, L. Tang, S. Wang, C. Lin, Y. Gong, L. M. Ni, H.-Y. Shum, and J. Guo. Think-on-graph: Deep and responsible reasoning of large language model on knowledge graph.International Conference on Learning Representations, 2024

    2024

  32. [32]

    K. Sun, Z. Zhao, H. Yang, J. Zhang, and G. Q. Huang. Curriculum engineering: Structured learning for large language models (LLMs) through curriculum based retrieval.IEEE Transactions on Industrial Informatics, 22(1):555–566, 2026

    2026

  33. [33]

    Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents

    Y. Talebirad and A. Nadiri. Multi-agent collaboration: Harnessing the power of intelligent LLM agents.arXiv preprint arXiv:2306.03314, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  34. [34]

    Tang et al

    L. Tang et al. Characteristics and health risks of volatile organic compounds emitted from a typical iron and steel industry in north china.Environ. Res., 212:113337, 2022

    2022

  35. [35]

    Taylor et al

    R. Taylor et al. Galactica: A large language model for science.arXiv, 2022

    2022

  36. [36]

    F. Wang, D. Shi, J. Aguilar, X. Cui, J. Jiang, L. Shen, and M. Li. Llm-kgmqa: large language model-augmented multi-hop question-answering system based on knowledge graph in medical field.KNOWLEDGE AND INFORMATION SYSTEMS, 67(8):6461–6503, 2025

    2025

  37. [37]

    H.-T. Wang, X. Bai, Z. Zheng, X. Zhang, R. Jin, H.-T. An, Z.-H. Xie, X.-L. Lv, and J.-R. Li. Chat-rfb: A flow battery chat system leveraging knowledge graphs and large language models.Digital Discovery, 5:1401–1410, 2026

    2026

  38. [38]

    Wang et al

    L. Wang et al. A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6):186345, 2024

    2024

  39. [39]

    Y. Wang, N. Lipka, R. A. Rossi, A. Siu, R. Zhang, and T. Derr. Knowledge graph prompting for multi-document question answering. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19206–19214, 2024

    2024

  40. [40]

    A. D. White et al. The future of chemistry is language.Nature Reviews Chemistry, 7(7):457–458, 2023

    2023

  41. [41]

    Wu et al

    Q. Wu et al. Autogen: Enabling next-gen LLM applications via multi-agent conversation.arXiv, 2023

    2023

  42. [42]

    Wu et al

    X. Wu et al. Augmented and programmatically optimized llm prompts reduce chemical hallucinations.Journal of Chemical Information and Modeling, 2025. Published online ahead of print

    2025

  43. [43]

    J. Xu, H. Zhang, H. Zhang, J. Lu, and G. Xiao. Chattf: A knowledge graph-enhanced intelligent q&a system for mitigating factuality hallucinations in traditional folklore.IEEE ACCESS, 12:162638–162650, 2024

    2024

  44. [44]

    S. Xu, L. Li, Z. Li, Y. Yao, F. Xu, Z. Chen, Q. Lu, and H. Tong. On the vulnerability of graph learning-based collaborative filtering.ACM Transactions on Information Systems, 41(4):Article 87, 2023

    2023

  45. [45]

    Yao et al

    S. Yao et al. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations, 2023

    2023

  46. [46]

    Zhang, L

    J. Zhang, L. Sang, Y. Xu, and H. Sun. Networked multiagent-based safe reinforcement learning for low-carbon demand management in distribution networks.IEEE Transactions on Sustainable Energy, 15(3):1528–1545, July 2024. 35/36

    2024

  47. [47]

    Zhang et al

    X. Zhang et al. Recent advances in the catalytic oxidation of volatile organic compounds: A review based on pollutant sorts and sources.Chemical Reviews, 116(6):3622–3673, 2016

    2016

  48. [48]

    Zhang, X

    Y. Zhang, X. Bo, Y. Zhao, C. P. Nielsen, J. Zhang, S. Zhang, X. Li, B. Zhao, G. Geng, and Q. Zhang. Benefits of current and future policies on emissions of china’s coal-fired power sector indicated by continuous emission monitoring. Environmental Pollution, 251:415–424, 2019

    2019

  49. [49]

    Zhang et al

    Y. Zhang et al. Emission characteristics and synergistic control technologies of volatile organic compounds in typical industrial processes.Journal of Hazardous Materials, 424:127645, 2022

    2022

  50. [50]

    Zheng and G

    P. Zheng and G. Xu. Expert systems: Grounding cross-disciplinary LLMs in reality.Resources Chemicals and Materials, 5(1):100164, 2026

    2026

  51. [51]

    Zheng et al

    Z. Zheng et al. Chatgpt in chemical research and education.ACS Central Science, 9(8):1459–1466, 2023

    2023

  52. [52]

    B. Zhou, X. Li, T. Liu, K. Xu, W. Liu, and J. Bao. Causalkgpt: Industrial structure causal knowledge-enhanced large language model for cause analysis of quality problems in aerospace product manufacturing.Advanced Engineering Informatics, 59:102333, 2024

    2024

  53. [53]

    Y. Zhu, X. Wang, J. Chen, S. Qiao, Y. Ou, Y. Yao, S. Deng, H. Chen, and N. Zhang. Llms for knowledge graph construction and reasoning: recent capabilities and future opportunities.WORLD WIDE WEB-INTERNET AND WEB INFORMATION SYSTEMS, 27(5):58, 2024. 36/36

    2024