arxiv: 2602.15353 · v2 · submitted 2026-02-17 · 💻 cs.CL · cs.AI

Recognition: no theorem link

NeuroSymActive: Differentiable Neural-Symbolic Reasoning with Active Exploration for Knowledge Graph Question Answering

Rong Fu , Yang Li , Zeyu Zhang , Jiekai Wu , Yaohua Liu , Shuaishuai Cao , Yangchen Zeng , Yuhang Zhang

show 2 more authors

Xiaojing Du Simon Fong

Authors on Pith no claims yet

Pith reviewed 2026-05-15 22:07 UTC · model grok-4.3

classification 💻 cs.CL cs.AI

keywords neural-symbolic reasoningknowledge graph question answeringdifferentiable symbolic modulesactive explorationMonte-Carlo path searchmulti-hop inferenceretrieval efficiency

0 comments

The pith

NeuroSymActive pairs differentiable neural-symbolic modules with value-guided Monte-Carlo exploration to answer multi-hop knowledge-graph questions more accurately and with fewer lookups than standard baselines.

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

The paper presents NeuroSymActive as a way to combine neural language models with symbolic graph reasoning for questions that need several hops of factual inference. It replaces brute-force retrieval with an active controller that uses value estimates to decide which paths to expand next. Soft-unification modules let the symbolic parts remain differentiable so gradients can adjust both the neural evaluator and the reasoning steps together. Experiments on common KGQA benchmarks show the approach matches or exceeds accuracy of retrieval-augmented models while cutting the number of graph accesses and model calls.

Core claim

NeuroSymActive couples soft-unification style symbolic modules with a neural path evaluator and a Monte-Carlo style exploration policy that prioritizes high-value path expansions, attaining strong answer accuracy on KGQA benchmarks while reducing the number of expensive graph lookups and model calls compared to common retrieval-augmented baselines.

What carries the argument

The active value-guided Monte-Carlo exploration controller that works with soft-unification symbolic modules and a neural path evaluator to select promising reasoning paths.

If this is right

The same modular design allows different symbolic reasoning components to be swapped in while keeping end-to-end gradient flow.
Fewer graph lookups make the method practical for knowledge bases that are expensive or rate-limited to access.
Value-guided selection focuses computation on high-reward paths, which directly lowers the total number of model evaluations needed.
The approach supports multi-hop queries without requiring the entire graph to be embedded in a single prompt.

Where Pith is reading between the lines

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

The same active-exploration pattern could be applied to other search-heavy tasks such as automated planning or theorem proving.
Because the controller is value-guided, it may remain effective even when the underlying knowledge graph is incomplete or contains noisy facts.
Scaling the framework to dynamic graphs that change over time would require only retraining the neural evaluator rather than redesigning the symbolic layer.

Load-bearing premise

That the soft-unification modules and the value-guided exploration policy will combine without hidden performance loss and will continue to work on graphs and question sets larger or different from those tested.

What would settle it

A controlled test on a new or larger knowledge graph where NeuroSymActive requires more graph lookups than a simple retrieval baseline while matching or falling below its accuracy would falsify the efficiency claim.

Figures

Figures reproduced from arXiv: 2602.15353 by Jiekai Wu, Rong Fu, Shuaishuai Cao, Simon Fong, Xiaojing Du, Yangchen Zeng, Yang Li, Yaohua Liu, Yuhang Zhang, Zeyu Zhang.

**Figure 1.** Figure 1: Architectural overview of the NeuroSymActive framework for knowledge graph question answering. The framework operates via a coupled dual-loop optimization process across three main stages: Stage 1: UncertaintyAware Active Retrieval, which utilizes a Bayesian head to model heteroscedastic uncertainty in hop prediction and a neural entropy predictor ηθ to estimate information gain (IG). This stage selective… view at source ↗

**Figure 2.** Figure 2: Reduction in error rates for each failure mode as annotation budget increases. Confidence intervals obtained [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

**Figure 3.** Figure 3: Composition of residual errors under strong active supervision. Components include KG incompleteness, [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

**Figure 4.** Figure 4: Accuracy versus annotation cost as controlled by uncertainty threshold [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗

**Figure 9.** Figure 9: Distribution of node expansions by tree depth and query hop count. Human query nodes marked separately. [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗

**Figure 5.** Figure 5: Marginal information gain across successive human queries within episodes. Shaded area denotes variance [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗

**Figure 10.** Figure 10: Effective branching factor versus visit count, stratified by uncertainty quartiles; dashed line shows the [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗

**Figure 6.** Figure 6: Accuracy under varying annotation budgets. NeuroSymActive outperforms uncertainty-agnostic baselines [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗

**Figure 7.** Figure 7: Normalized loss-term magnitudes across training iterations. Shaded bands show seed-wise variation. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗

**Figure 8.** Figure 8: Evolution of rule confidences in DILL, grouped by rule category. Solid lines denote means; shading indicates [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗

read the original abstract

Large pretrained language models and neural reasoning systems have advanced many natural language tasks, yet they remain challenged by knowledge-intensive queries that require precise, structured multi-hop inference. Knowledge graphs provide a compact symbolic substrate for factual grounding, but integrating graph structure with neural models is nontrivial: naively embedding graph facts into prompts leads to inefficiency and fragility, while purely symbolic or search-heavy approaches can be costly in retrievals and lack gradient-based refinement. We introduce NeuroSymActive, a modular framework that combines a differentiable neural-symbolic reasoning layer with an active, value-guided exploration controller for Knowledge Graph Question Answering. The method couples soft-unification style symbolic modules with a neural path evaluator and a Monte-Carlo style exploration policy that prioritizes high-value path expansions. Empirical results on standard KGQA benchmarks show that NeuroSymActive attains strong answer accuracy while reducing the number of expensive graph lookups and model calls compared to common retrieval-augmented baselines.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

NeuroSymActive couples soft-unification modules with value-guided Monte-Carlo exploration to cut graph lookups while holding accuracy on standard KGQA benchmarks.

read the letter

The main point is that this framework integrates differentiable symbolic reasoning with an active exploration policy, and the reported experiments show it maintains answer quality while lowering the number of expensive lookups and model calls versus retrieval baselines. The modular design lets gradients flow through the soft-unification steps and uses the neural evaluator to score paths, so the Monte-Carlo policy can focus expansions on high-value branches instead of exhaustive search. That combination is what produces the efficiency numbers on the benchmarks they tested. The results look grounded enough to support the central claim, with the active component appearing to deliver measurable savings without obvious collapse in accuracy. One soft spot is that everything is evaluated on the usual KGQA datasets, so it is still open how the value estimates hold up on noisier or much larger graphs where path scoring could degrade. A few more targeted ablations on the exploration policy versus simpler beam search would also make the contribution sharper. This is the sort of paper that matters to groups building hybrid systems for factual multi-hop queries who already care about compute cost. The thinking is straightforward and the empirical comparisons are usable, so it is worth sending out for real referee feedback rather than desk rejection.

Referee Report

0 major / 3 minor

Summary. The paper introduces NeuroSymActive, a modular framework for Knowledge Graph Question Answering that combines a differentiable neural-symbolic reasoning layer (using soft-unification style symbolic modules and a neural path evaluator) with an active value-guided Monte-Carlo exploration controller. It claims to deliver strong answer accuracy on standard KGQA benchmarks while reducing the number of expensive graph lookups and model calls relative to common retrieval-augmented baselines.

Significance. If the empirical results hold, the work offers a practical advance in neural-symbolic KGQA by showing how targeted exploration can maintain accuracy with lower retrieval and inference cost; the differentiable components enable end-to-end refinement, which is a clear methodological strength over purely symbolic or prompt-only approaches.

minor comments (3)

[§4] §4 (Method): the integration of the soft-unification modules with the value-guided policy is described at a high level; adding a short pseudocode block or explicit loss formulation would improve reproducibility.
[Table 2] Table 2: the reported lookup reductions lack error bars or significance tests; including these would strengthen the efficiency claims.
[§5.3] §5.3 (Ablations): the contribution of the active exploration component versus a greedy baseline is shown, but the paper should explicitly state whether the neural path evaluator is frozen or jointly trained in each ablation setting.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary of NeuroSymActive and the recommendation for minor revision. The recognition of the framework's ability to maintain accuracy with reduced retrieval and inference costs, along with the value of its differentiable components, is appreciated. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents NeuroSymActive as a modular combination of existing neural-symbolic components (soft-unification modules, neural path evaluator) with a value-guided Monte-Carlo exploration policy. No equations or claims in the abstract or described framework reduce by construction to fitted parameters or self-citations; performance numbers are reported as empirical outcomes on standard KGQA benchmarks rather than tautological re-expressions of inputs. The derivation chain relies on integration of prior ideas with new controller logic, which remains externally testable and does not exhibit self-definitional, fitted-prediction, or uniqueness-imported circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; all technical details remain at the level of named components without equations or fitting procedures.

pith-pipeline@v0.9.0 · 5494 in / 1106 out tokens · 70186 ms · 2026-05-15T22:07:51.145085+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

74 extracted references · 74 canonical work pages · 3 internal anchors

[1]

Retrieval as attention: End-to-end learning of retrieval and reading within a single transformer

Zhengbao Jiang, Luyu Gao, Zhiruo Wang, Jun Araki, Haibo Ding, Jamie Callan, and Graham Neubig. Retrieval as attention: End-to-end learning of retrieval and reading within a single transformer. InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2336–2349, 2022

work page 2022
[2]

Self-retrieval: End-to-end information retrieval with one large language model.Advances in Neural Information Processing Systems, 37:63510–63533, 2024

Qiaoyu Tang, Jiawei Chen, Zhuoqun Li, Bowen Yu, Yaojie Lu, Haiyang Yu, Hongyu Lin, Fei Huang, Ben He, Xianpei Han, et al. Self-retrieval: End-to-end information retrieval with one large language model.Advances in Neural Information Processing Systems, 37:63510–63533, 2024

work page 2024
[3]

LightRAG: Simple and Fast Retrieval-Augmented Generation

Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang. Lightrag: Simple and fast retrieval-augmented generation.arXiv preprint arXiv:2410.05779, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024
[4]

Learning from interpretation transition using differentiable logic programming semantics.Machine Learning, 111(1):123–145, 2022

Kun Gao, Hanpin Wang, Yongzhi Cao, and Katsumi Inoue. Learning from interpretation transition using differentiable logic programming semantics.Machine Learning, 111(1):123–145, 2022

work page 2022
[5]

Soft-unification in deep probabilistic logic.Advances in Neural Information Processing Systems, 36:60804–60820, 2023

Jaron Maene and Luc De Raedt. Soft-unification in deep probabilistic logic.Advances in Neural Information Processing Systems, 36:60804–60820, 2023

work page 2023
[6]

Complex query answering on eventuality knowledge graph with implicit logical constraints.Advances in Neural Information Processing Systems, 36: 30534–30553, 2023

Jiaxin Bai, Xin Liu, Weiqi Wang, Chen Luo, and Yangqiu Song. Complex query answering on eventuality knowledge graph with implicit logical constraints.Advances in Neural Information Processing Systems, 36: 30534–30553, 2023

work page 2023
[7]

Beamqa: Multi-hop knowledge graph question answering with sequence-to-sequence prediction and beam search

Farah Atif, Ola El Khatib, and Djellel Difallah. Beamqa: Multi-hop knowledge graph question answering with sequence-to-sequence prediction and beam search. InProceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 781–790, 2023

work page 2023
[8]

Smartquery: An active learning framework for graph neural networks through hybrid uncertainty reduction

Xiaoting Li, Yuhang Wu, Vineeth Rakesh, Yusan Lin, Hao Yang, and Fei Wang. Smartquery: An active learning framework for graph neural networks through hybrid uncertainty reduction. InProceedings of the 31st ACM International Conference on Information & Knowledge Management, pages 4199–4203, 2022

work page 2022
[9]

Neuroescape: Ordered escape routing via monte-carlo tree search and neural network

Zhiyang Chen, Tsung-Yi Ho, Ulf Schlichtmann, Datao Chen, Mingyu Liu, Hailong Yao, and Xia Yin. Neuroescape: Ordered escape routing via monte-carlo tree search and neural network. In2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD), pages 01–09. IEEE, 2023

work page 2023
[10]

A critical review of inductive logic programming techniques for explainable ai.IEEE transactions on neural networks and learning systems, 35(8):10220–10236, 2023

Zheng Zhang, Levent Yilmaz, and Bo Liu. A critical review of inductive logic programming techniques for explainable ai.IEEE transactions on neural networks and learning systems, 35(8):10220–10236, 2023. 19 NeuroSymActive

work page 2023
[11]

Deeplogic: Joint learning of neural perception and logical reasoning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(4): 4321–4334, 2022

Xuguang Duan, Xin Wang, Peilin Zhao, Guangyao Shen, and Wenwu Zhu. Deeplogic: Joint learning of neural perception and logical reasoning.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(4): 4321–4334, 2022

work page 2022
[12]

Logicenn: A neural based knowledge graphs embedding model with logical rules.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(6):7050–7062, 2021

Mojtaba Nayyeri, Chengjin Xu, Mirza Mohtashim Alam, Jens Lehmann, and Hamed Shariat Yazdi. Logicenn: A neural based knowledge graphs embedding model with logical rules.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(6):7050–7062, 2021

work page 2021
[13]

Neupsl: Neural probabilistic soft logic.arXiv preprint arXiv:2205.14268, 2022

Connor Pryor, Charles Dickens, Eriq Augustine, Alon Albalak, William Wang, and Lise Getoor. Neupsl: Neural probabilistic soft logic.arXiv preprint arXiv:2205.14268, 2022

work page arXiv 2022
[14]

Differentiable inductive logic programming for structured examples

Hikaru Shindo, Masaaki Nishino, and Akihiro Yamamoto. Differentiable inductive logic programming for structured examples. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 5034–5041, 2021

work page 2021
[15]

α ilp: thinking visual scenes as differentiable logic programs.Machine Learning, 112(5):1465–1497, 2023

Hikaru Shindo, Viktor Pfanschilling, Devendra Singh Dhami, and Kristian Kersting. α ilp: thinking visual scenes as differentiable logic programs.Machine Learning, 112(5):1465–1497, 2023

work page 2023
[16]

Learning interpretable differentiable logic networks.IEEE Transactions on Circuits and Systems for Artificial Intelligence, 2024

Chang Yue and Niraj K Jha. Learning interpretable differentiable logic networks.IEEE Transactions on Circuits and Systems for Artificial Intelligence, 2024

work page 2024
[17]

Differentiable neuro-symbolic reasoning on large-scale knowledge graphs.Advances in Neural Information Processing Systems, 36:28139–28154, 2023

Chen Shengyuan, Yunfeng Cai, Huang Fang, Xiao Huang, and Mingming Sun. Differentiable neuro-symbolic reasoning on large-scale knowledge graphs.Advances in Neural Information Processing Systems, 36:28139–28154, 2023

work page 2023
[18]

Mote-nas: Multi- objective training-based estimate for efficient neural architecture search.Advances in Neural Information Processing Systems, 37:100845–100869, 2024

Yuming Zhang, Jun Hsieh, Xin Li, Ming-Ching Chang, Chun-Chieh Lee, and Kuo-Chin Fan. Mote-nas: Multi- objective training-based estimate for efficient neural architecture search.Advances in Neural Information Processing Systems, 37:100845–100869, 2024

work page 2024
[19]

Neuro-symbolic artificial intelligence: a survey.Neural Computing and Applications, 36(21):12809–12844, 2024

Bikram Pratim Bhuyan, Amar Ramdane-Cherif, Ravi Tomar, and TP Singh. Neuro-symbolic artificial intelligence: a survey.Neural Computing and Applications, 36(21):12809–12844, 2024

work page 2024
[20]

A review of neuro-symbolic ai integrating reasoning and learning for advanced cognitive systems.Intelligent Systems with Applications, page 200541, 2025

Uzma Nawaz, Mufti Anees-ur Rahaman, and Zubair Saeed. A review of neuro-symbolic ai integrating reasoning and learning for advanced cognitive systems.Intelligent Systems with Applications, page 200541, 2025

work page 2025
[21]

Human-in-the-loop machine learning: a state of the art.Artificial Intelligence Review, 56 (4):3005–3054, 2023

Eduardo Mosqueira-Rey, Elena Hernández-Pereira, David Alonso-Ríos, José Bobes-Bascarán, and Ángel Fernández-Leal. Human-in-the-loop machine learning: a state of the art.Artificial Intelligence Review, 56 (4):3005–3054, 2023

work page 2023
[22]

Efficient human-in-the-loop active learning: A novel framework for data labeling in ai systems.arXiv preprint arXiv:2501.00277, 2024

Yiran Huang, Jian-Feng Yang, and Haoda Fu. Efficient human-in-the-loop active learning: A novel framework for data labeling in ai systems.arXiv preprint arXiv:2501.00277, 2024

work page arXiv 2024
[23]

Active learning framework for improving knowledge graph accuracy.IEEE Access, 2025

Donghyun Kim, Hyeongjun Yang, Seokju Hwang, Kyuhwan Yeom, Midan Shim, and Kyong-Ho Lee. Active learning framework for improving knowledge graph accuracy.IEEE Access, 2025

work page 2025
[24]

Freeson: Retriever-free retrieval-augmented reasoning via corpus-traversing mcts.arXiv preprint arXiv:2505.16409, 2025

Chaeeun Kim and Seungone Kim. Freeson: Retriever-free retrieval-augmented reasoning via corpus-traversing mcts.arXiv preprint arXiv:2505.16409, 2025

work page arXiv 2025
[25]

Few-shot preference learning for human-in-the-loop rl

Donald Joseph Hejna III and Dorsa Sadigh. Few-shot preference learning for human-in-the-loop rl. InConference on Robot Learning, pages 2014–2025. PMLR, 2023

work page 2014
[26]

Nc-alg: Graph-based active learning under noisy crowd

Wentao Zhang, Yexin Wang, Zhenbang You, Yang Li, Gang Cao, Zhi Yang, and Bin Cui. Nc-alg: Graph-based active learning under noisy crowd. In2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 2681–2694. IEEE, 2024

work page 2024
[27]

A comparison of uncertainty quantification methods for active learning in image classification

Alice Hein, Stefan Röhrl, Thea Grobel, Manuel Lengl, Nawal Hafez, Martin Knopp, Christian Klenk, Dominik Heim, Oliver Hayden, and Klaus Diepold. A comparison of uncertainty quantification methods for active learning in image classification. In2022 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE, 2022

work page 2022
[28]

Epistemic uncertainty quantification for pre-trained neural networks

Hanjing Wang and Qiang Ji. Epistemic uncertainty quantification for pre-trained neural networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11052–11061, 2024. 20 NeuroSymActive

work page 2024
[29]

Effective bayesian heteroscedastic regression with deep neural networks.Advances in Neural Information Processing Systems, 36:53996–54019, 2023

Alexander Immer, Emanuele Palumbo, Alexander Marx, and Julia V ogt. Effective bayesian heteroscedastic regression with deep neural networks.Advances in Neural Information Processing Systems, 36:53996–54019, 2023

work page 2023
[30]

On the pitfalls of heteroscedastic uncertainty estimation with probabilistic neural networks.arXiv preprint arXiv:2203.09168, 2022

Maximilian Seitzer, Arash Tavakoli, Dimitrije Antic, and Georg Martius. On the pitfalls of heteroscedastic uncertainty estimation with probabilistic neural networks.arXiv preprint arXiv:2203.09168, 2022

work page arXiv 2022
[31]

Bayesian entropy neural networks for physics- aware prediction.arXiv preprint arXiv:2407.01015, 2024

Rahul Rathnakumar, Jiayu Huang, Hao Yan, and Yongming Liu. Bayesian entropy neural networks for physics- aware prediction.arXiv preprint arXiv:2407.01015, 2024

work page arXiv 2024
[32]

Uncertainty-aware relational graph neural network for few-shot knowledge graph completion.arXiv preprint arXiv:2403.04521, 2024

Qian Li, Shu Guo, Yinjia Chen, Cheng Ji, Jiawei Sheng, and Jianxin Li. Uncertainty-aware relational graph neural network for few-shot knowledge graph completion.arXiv preprint arXiv:2403.04521, 2024

work page arXiv 2024
[33]

Active learning with fully bayesian neural networks for discontinuous and nonstationary data

Maxim Ziatdinov. Active learning with fully bayesian neural networks for discontinuous and nonstationary data. arXiv preprint arXiv:2405.09817, 2024

work page arXiv 2024
[34]

Graph neural networks for multi-robot active information acquisition.arXiv preprint arXiv:2209.12091, 2022

Mariliza Tzes, Nikolaos Bousias, Evangelos Chatzipantazis, and George J Pappas. Graph neural networks for multi-robot active information acquisition.arXiv preprint arXiv:2209.12091, 2022

work page arXiv 2022
[35]

Rag-ddr: Optimizing retrieval-augmented generation using differentiable data rewards.arXiv preprint arXiv:2410.13509, 2024

Xinze Li, Sen Mei, Zhenghao Liu, Yukun Yan, Shuo Wang, Shi Yu, Zheni Zeng, Hao Chen, Ge Yu, Zhiyuan Liu, et al. Rag-ddr: Optimizing retrieval-augmented generation using differentiable data rewards.arXiv preprint arXiv:2410.13509, 2024

work page arXiv 2024
[36]

Knowledge graph prompting for multi-document question answering

Yu Wang, Nedim Lipka, Ryan A Rossi, Alexa Siu, Ruiyi Zhang, and Tyler Derr. Knowledge graph prompting for multi-document question answering. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 19206–19214, 2024

work page 2024
[37]

Retrieval and reasoning on kgs: Integrate knowledge graphs into large language models for complex question answering

Yixin Ji, Kaixin Wu, Juntao Li, Wei Chen, Mingjie Zhong, Xu Jia, and Min Zhang. Retrieval and reasoning on kgs: Integrate knowledge graphs into large language models for complex question answering. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 7598–7610, 2024

work page 2024
[38]

Simple is effective: The roles of graphs and large language models in knowledge-graph-based retrieval-augmented generation.arXiv preprint arXiv:2410.20724, 2024

Mufei Li, Siqi Miao, and Pan Li. Simple is effective: The roles of graphs and large language models in knowledge-graph-based retrieval-augmented generation.arXiv preprint arXiv:2410.20724, 2024

work page arXiv 2024
[39]

D-rag: Differentiable retrieval-augmented generation for knowledge graph question answering

Guangze Gao, Zixuan Li, Chunfeng Yuan, Jiawei Li, Wu Jianzhuo, Yuehao Zhang, Xiaolong Jin, Bing Li, and Weiming Hu. D-rag: Differentiable retrieval-augmented generation for knowledge graph question answering. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 35386–35405, 2025

work page 2025
[40]

Rekg-mcts: Reinforcing llm reasoning on knowledge graphs via training-free monte carlo tree search

Xiaozhuang Song, Shufei Zhang, and Tianshu Yu. Rekg-mcts: Reinforcing llm reasoning on knowledge graphs via training-free monte carlo tree search. InFindings of the Association for Computational Linguistics: ACL 2025, pages 9288–9306, 2025

work page 2025
[41]

Interpretable contrastive monte carlo tree search reasoning.arXiv preprint arXiv:2410.01707, 2024

Zitian Gao, Boye Niu, Xuzheng He, Haotian Xu, Hongzhang Liu, Aiwei Liu, Xuming Hu, and Lijie Wen. Interpretable contrastive monte carlo tree search reasoning.arXiv preprint arXiv:2410.01707, 2024

work page arXiv 2024
[42]

Exact: Teaching ai agents to explore with reflective-mcts and exploratory learning.arXiv preprint arXiv:2410.02052, 2024

Xiao Yu, Baolin Peng, Vineeth Vajipey, Hao Cheng, Michel Galley, Jianfeng Gao, and Zhou Yu. Exact: Teaching ai agents to explore with reflective-mcts and exploratory learning.arXiv preprint arXiv:2410.02052, 2024

work page arXiv 2024
[43]

Improving discrete optimisation via decoupled straight-through gumbel-softmax.arXiv preprint arXiv:2410.13331, 2024

Rushi Shah, Mingyuan Yan, Michael Curtis Mozer, and Dianbo Liu. Improving discrete optimisation via decoupled straight-through gumbel-softmax.arXiv preprint arXiv:2410.13331, 2024

work page arXiv 2024
[44]

Gumbel-softmax based graph convolution network approach for community detection.International Journal of Information Technology, 15(6):3063–3070, 2023

Laxmi Chaudhary and Buddha Singh. Gumbel-softmax based graph convolution network approach for community detection.International Journal of Information Technology, 15(6):3063–3070, 2023

work page 2023
[45]

Path spuriousness-aware reinforcement learning for multi-hop knowledge graph reasoning

Chunyang Jiang, Tianchen Zhu, Haoyi Zhou, Chang Liu, Ting Deng, Chunming Hu, and Jianxin Li. Path spuriousness-aware reinforcement learning for multi-hop knowledge graph reasoning. InProceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 3181–3192, 2023

work page 2023
[46]

Rpr-kgqa: Relational path reasoning for multi-hop question answering with knowledge graph

Long Zhao, Yin Xu, Yanyan Wang, Zhengyi Chen, Youzhi Huang, and Qiangzhong Feng. Rpr-kgqa: Relational path reasoning for multi-hop question answering with knowledge graph. InProceedings of the 2024 International Conference on Computer and Multimedia Technology, pages 596–600, 2024. 21 NeuroSymActive

work page 2024
[47]

Hierarchy-aware multi-hop question answering over knowledge graphs

Junnan Dong, Qinggang Zhang, Xiao Huang, Keyu Duan, Qiaoyu Tan, and Zhimeng Jiang. Hierarchy-aware multi-hop question answering over knowledge graphs. InProceedings of the ACM web conference 2023, pages 2519–2527, 2023

work page 2023
[48]

Tooltree: Efficient llm agent tool planning via dual-feedback monte carlo tree search and bidirectional pruning.arXiv preprint arXiv:2603.12740, 2026

Shuo Yang, Soyeon Caren Han, Yihao Ding, Shuhe Wang, and Eduard Hoy. Tooltree: Efficient llm agent tool planning via dual-feedback monte carlo tree search and bidirectional pruning.arXiv preprint arXiv:2603.12740, 2026

work page arXiv 2026
[49]

Evotool: Self-evolving tool-use policy optimization in llm agents via blame-aware mutation and diversity-aware selection

Shuo Yang, Soyeon Caren Han, Xueqi Ma, Yan Li, Mohammad Reza Ghasemi Madani, and Eduard Hovy. Evotool: Self-evolving tool-use policy optimization in llm agents via blame-aware mutation and diversity-aware selection. arXiv preprint arXiv:2603.04900, 2026

work page arXiv 2026
[50]

Subgraph retrieval enhanced model for multi-hop knowledge base question answering.arXiv preprint arXiv:2202.13296, 2022

Jing Zhang, Xiaokang Zhang, Jifan Yu, Jian Tang, Jie Tang, Cuiping Li, and Hong Chen. Subgraph retrieval enhanced model for multi-hop knowledge base question answering.arXiv preprint arXiv:2202.13296, 2022

work page arXiv 2022
[51]

Multi-hop knowledge graph reasoning in few-shot scenarios.IEEE Transactions on Knowledge and Data Engineering, 36(4):1713–1727, 2023

Shangfei Zheng, Wei Chen, Weiqing Wang, Pengpeng Zhao, Hongzhi Yin, and Lei Zhao. Multi-hop knowledge graph reasoning in few-shot scenarios.IEEE Transactions on Knowledge and Data Engineering, 36(4):1713–1727, 2023

work page 2023
[52]

Pullnet: Open domain question answering with iterative retrieval on knowledge bases and text

Haitian Sun, Tania Bedrax-Weiss, and William Cohen. Pullnet: Open domain question answering with iterative retrieval on knowledge bases and text. InProceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 2380–2390, 2019

work page 2019
[53]

Open domain question answering using early fusion of knowledge bases and text

Haitian Sun, Bhuwan Dhingra, Manzil Zaheer, Kathryn Mazaitis, Ruslan Salakhutdinov, and William Cohen. Open domain question answering using early fusion of knowledge bases and text. InProceedings of the 2018 conference on empirical methods in natural language processing, pages 4231–4242, 2018

work page 2018
[54]

Transfernet: An effective and transparent framework for multi-hop question answering over relation graph.arXiv preprint arXiv:2104.07302, 2021

Jiaxin Shi, Shulin Cao, Lei Hou, Juanzi Li, and Hanwang Zhang. Transfernet: An effective and transparent framework for multi-hop question answering over relation graph.arXiv preprint arXiv:2104.07302, 2021

work page arXiv 2021
[55]

Interactive-kbqa: Multi-turn interactions for knowledge base question answering with large language models.arXiv preprint arXiv:2402.15131, 2024

Guanming Xiong, Junwei Bao, and Wen Zhao. Interactive-kbqa: Multi-turn interactions for knowledge base question answering with large language models.arXiv preprint arXiv:2402.15131, 2024

work page arXiv 2024
[56]

Chain-of-question: A progressive question decomposition approach for complex knowledge base question answering

Peng Yixing, Quan Wang, Licheng Zhang, Yi Liu, and Zhendong Mao. Chain-of-question: A progressive question decomposition approach for complex knowledge base question answering. InFindings of the Association for Computational Linguistics ACL 2024, pages 4763–4776, 2024

work page 2024
[57]

Chatgpt versus traditional question answering for knowledge graphs: Current status and future directions towards knowledge graph chatbots.arXiv preprint arXiv:2302.06466, 2023

Reham Omar, Omij Mangukiya, Panos Kalnis, and Essam Mansour. Chatgpt versus traditional question answering for knowledge graphs: Current status and future directions towards knowledge graph chatbots.arXiv preprint arXiv:2302.06466, 2023

work page arXiv 2023
[58]

Knowprefix-tuning: A two-stage prefix-tuning framework for knowledge-grounded dialogue generation

Jiaqi Bai, Zhao Yan, Ze Yang, Jian Yang, Xinnian Liang, Hongcheng Guo, and Zhoujun Li. Knowprefix-tuning: A two-stage prefix-tuning framework for knowledge-grounded dialogue generation. InJoint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 525–542. Springer, 2023

work page 2023
[59]

Infoprompt: Information-theoretic soft prompt tuning for natural language understanding.Advances in neural information processing systems, 36:61060–61084, 2023

Junda Wu, Tong Yu, Rui Wang, Zhao Song, Ruiyi Zhang, Handong Zhao, Chaochao Lu, Shuai Li, and Ricardo Henao. Infoprompt: Information-theoretic soft prompt tuning for natural language understanding.Advances in neural information processing systems, 36:61060–61084, 2023

work page 2023
[60]

Lifelong knowledge editing for llms with retrieval-augmented continuous prompt learning

Qizhou Chen, Taolin Zhang, Xiaofeng He, Dongyang Li, Chengyu Wang, Longtao Huang, et al. Lifelong knowledge editing for llms with retrieval-augmented continuous prompt learning. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 13565–13580, 2024

work page 2024
[61]

Uprise: Universal prompt retrieval for improving zero-shot evaluation

Daixuan Cheng, Shaohan Huang, Junyu Bi, Yuefeng Zhan, Jianfeng Liu, Yujing Wang, Hao Sun, Furu Wei, Weiwei Deng, and Qi Zhang. Uprise: Universal prompt retrieval for improving zero-shot evaluation. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 12318–12337, 2023

work page 2023
[62]

Automatic optimization method for database indexing by integrating monte carlo tree search and graph neural network.Procedia Computer Science, 262:831–839, 2025

Shanyue Wan. Automatic optimization method for database indexing by integrating monte carlo tree search and graph neural network.Procedia Computer Science, 262:831–839, 2025

work page 2025
[63]

The value of semantic parse labeling for knowledge base question answering

Wen-tau Yih, Matthew Richardson, Christopher Meek, Ming-Wei Chang, and Jina Suh. The value of semantic parse labeling for knowledge base question answering. InProceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 201–206, 2016. 22 NeuroSymActive

work page 2016
[64]

The Web as a Knowledge-base for Answering Complex Questions

Alon Talmor and Jonathan Berant. The web as a knowledge-base for answering complex questions.arXiv preprint arXiv:1803.06643, 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[65]

Key-value memory networks for directly reading documents (emnlp16).arXiv, 2016

Alexander H Miller, Adam Fisch, Jesse Dodge, Amir-Hossein Karimi, Antoine Bordes, and Jason Weston. Key-value memory networks for directly reading documents (emnlp16).arXiv, 2016

work page 2016
[66]

Improving multi-hop question answering over knowledge graphs using knowledge base embeddings

Apoorv Saxena, Aditay Tripathi, and Partha Talukdar. Improving multi-hop question answering over knowledge graphs using knowledge base embeddings. InProceedings of the 58th annual meeting of the association for computational linguistics, pages 4498–4507, 2020

work page 2020
[67]

Improving multi-hop knowledge base question answering by learning intermediate supervision signals

Gaole He, Yunshi Lan, Jing Jiang, Wayne Xin Zhao, and Ji-Rong Wen. Improving multi-hop knowledge base question answering by learning intermediate supervision signals. InProceedings of the 14th ACM international conference on web search and data mining, pages 553–561, 2021

work page 2021
[68]

Sequence-to-sequence knowledge graph completion and question answering.arXiv preprint arXiv:2203.10321, 2022

Apoorv Saxena, Adrian Kochsiek, and Rainer Gemulla. Sequence-to-sequence knowledge graph completion and question answering.arXiv preprint arXiv:2203.10321, 2022

work page arXiv 2022
[69]

Unikgqa: Unified retrieval and reasoning for solving multi-hop question answering over knowledge graph.arXiv preprint arXiv:2212.00959, 2022

Jinhao Jiang, Kun Zhou, Wayne Xin Zhao, and Ji-Rong Wen. Unikgqa: Unified retrieval and reasoning for solving multi-hop question answering over knowledge graph.arXiv preprint arXiv:2212.00959, 2022

work page arXiv 2022
[70]

Think-on-graph: Deep and responsible reasoning of large language model on knowledge graph.arXiv preprint arXiv:2307.07697, 2023

Jiashuo Sun, Chengjin Xu, Lumingyuan Tang, Saizhuo Wang, Chen Lin, Yeyun Gong, Lionel M Ni, Heung-Yeung Shum, and Jian Guo. Think-on-graph: Deep and responsible reasoning of large language model on knowledge graph.arXiv preprint arXiv:2307.07697, 2023

work page arXiv 2023
[71]

Structgpt: A general framework for large language model to reason over structured data.arXiv preprint arXiv:2305.09645, 2023

Jinhao Jiang, Kun Zhou, Zican Dong, Keming Ye, Wayne Xin Zhao, and Ji-Rong Wen. Structgpt: A general framework for large language model to reason over structured data.arXiv preprint arXiv:2305.09645, 2023

work page arXiv 2023
[72]

AgentBench: Evaluating LLMs as Agents

Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, et al. Agentbench: Evaluating llms as agents.arXiv preprint arXiv:2308.03688, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[73]

Knowledgenavigator: Leveraging large language models for enhanced reasoning over knowledge graph.Complex & Intelligent Systems, 10(5):7063–7076, 2024

Tiezheng Guo, Qingwen Yang, Chen Wang, Yanyi Liu, Pan Li, Jiawei Tang, Dapeng Li, and Yingyou Wen. Knowledgenavigator: Leveraging large language models for enhanced reasoning over knowledge graph.Complex & Intelligent Systems, 10(5):7063–7076, 2024

work page 2024
[74]

Lightprof: A lightweight reasoning framework for large language model on knowledge graph

Tu Ao, Yanhua Yu, Yuling Wang, Yang Deng, Zirui Guo, Liang Pang, Pinghui Wang, Tat-Seng Chua, Xiao Zhang, and Zhen Cai. Lightprof: A lightweight reasoning framework for large language model on knowledge graph. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 23424–23432, 2025. A Theoretical Analysis This section collects ...

work page 2025