Neural-Symbolic Message Passing with Dynamic Pruning
Keywords: knowledge graph, complex query answering, neural and symbolic, message passing
TL;DR: We propose a neural-symbolic message passing framework to solve complex query answering over knowledge graphs.
Abstract: Complex Query Answering (CQA) over Knowledge Graphs (KGs) is a fundamental yet challenging task. Given that KGs are usually incomplete, the CQA models not only need to execute logical operators, but aslo need to leverage observed knowledge to predict the missing one. Recently, a line of message-passing-based research has been proposed to re-use pre-trained neural link predictors to solve CQA. However, they perform unsatisfactorily on negative queries and fail to address the unnecessary noisy messages between variable nodes in the query graph. Moreover, like most neural CQA models, these message passing models offer little interpretability and require complex query data and resource-intensive training. In this paper, we propose a Neural-Symbolic Message Passing framework (NSMP), which integrates neural and symbolic reasoning. By re-using a simple pre-trained neural link predictor, NSMP generalizes to complex queries based on fuzzy logic theory, without requiring training on complex query datasets, while providing interpretable answers. Furthermore, we introduce an effective dynamic pruning strategy to filter out noisy messages between variable nodes during message passing. Empirically, our model demonstrates strong performance and offers efficient inference. Our code can be found at https://anonymous.4open.science/r/NSMP.
Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2632
Loading