A graph neural network (GNN)-based adaptive knowledge reasoning framework (CASS) was proposed to address the limitations of extant GNN-based knowledge graph reasoning, such as the explosive growth of entity propagation scale, loss of semantic associations, and sparsity of relational data. First, a connection-oriented semantic correlation sampling strategy was designed to filter redundant entities and retain potential targets through adaptive propagation paths. This strategy employed an improved progressive propagation mechanism to maintain linear complexity, and integrated dynamic sampling distribution with iterative elimination strategy to prune invalid nodes. Second, a masked graph-based contrastive learning paradigm was introduced. It utilized a learnable masking function to locate and mask the subgraph structures of critical nodes to reconstruct self-supervised signals, thereby enhancing the reasoning robustness of the model in sparse scenarios. Finally, experiments on public datasets WN18RR, NELL-995, and YAGO3-10 demonstrated that CASS achieved significant performance improvements in transductive reasoning tasks. In particular, on the WN18RR dataset, its mean reciprocal rank (MRR) yielded a 1.8 percentage point improvement over state-of-the-art baseline models.
PanS R, LuoL H, WangY F, et al. Unifying large language models and knowledge graphs: A roadmap[J]. IEEE Transactions on Knowledge and Data Engineering, 2024, 36(7): 3580-3599.
[2]
SunZ Q, DengZ H, NieJ Y, et al. RotatE: Knowledge graph embedding by relational rotation in complex space[PP/OL]. arXiv (2019-02-26)[2026-03-01].
[3]
SchlichtkrullM, KipfT N, BloemP, et al. Modeling relational data with graph convolutional networks[C]//European semantic web conference. Cham: Springer International Publishing, 2018: 593-607.
[4]
YeZ, KumarY J, SingG O, et al. A comprehensive survey of graph neural networks for knowledge graphs[J]. IEEE Access, 2022, 10: 75729-75741.
OonoK, SuzukiT. Graph neural networks exponentially lose expressive power for node classification[PP/OL]. V5. arXiv (2021-01-06) [2026-03-01].
[7]
ZhuZ, ZhangZ, XhonneuxL P, et al. Neural bellman-ford networks: A general graph neural network framework for link prediction[J]. Advances in neural information processing systems, 2021, 34: 29476-29490.
[8]
ZhangY Q, YaoQ M. Knowledge graph reasoning with relational digraph[C]//Proceedings of the ACM Web Conference 2022. Virtual Event, Lyon France. ACM, 2022: 912-924.
[9]
BordesA, UsunierN, Garcia-DuránA, et al. Translating embeddings for modeling multi-relational data[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2. ACM, 2013: 2787-2795.
[10]
YangB S, YihW T, HeX D, et al. Embedding entities and relations for learning and inference in knowledge bases[PP/OL]. V4. arXiv (2015-08-29)[ 2026-03-01].
[11]
XiongW H, HoangT, WangW Y. DeepPath: A reinforcement learning method for knowledge graph reasoning[C]//Proceedings of the 2017 Conference on Empirical Methods in NaturalLanguage Processing. Copenhagen, Denmark. ACL, 2017: 564-573.
[12]
HamiltonW L, YingR, LeskovecJ. Inductive representation learning on large graphs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. ACM, 2017: 1025-1035.
[13]
WangX, JiH Y, ShiC, et al. Heterogeneous graph attention network[C]//The World Wide Web Conference. ACM, 2019: 2022-2032.
[14]
WangZ, MaS X, WangK W, et al. Rule-guided graph neural networks for explainable knowledge graph reasoning[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2025, 39(12): 12784-12791.
[15]
WiltC, ThayerJ, RumlW. A comparison of greedy search algorithms[J]. Proceedings of the International Symposium on Combinatorial Search, 2010, 1(1): 129-136.
[16]
KoolW, Van HoofH, WellingM. Stochastic beams and where to find them: The gumbel-top-k trick for sampling sequences without replacement[C]//International conference on machine learning. PMLR, 2019: 3499-3508.
[17]
XieS M, ErmonS. Reparameterizable subset sampling via continuous relaxations[C]//Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence. August 10-16, 2019. Macao, China. International Joint Conferences on Artificial Intelligence Organization, 2019: 3919-3925.
BengioY, Le´onardNicholas, CourvilleA.Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation[EB/OL].Computer Science, 2013.DOI:10.48550/arXiv.1305.2982 .
[20]
KingmaD, BaJ. Adam: A Method for Stochastic Optimization[EB/OL].Computer Science, 2014.DOI:10.48550/arXiv.1412.6980 .