Experiments show that RelShap produces explanations that are more faithful to the data-generating process, correctly identifying the dominant feature in controlled settings where existing methods, including Conditional SHAP and ManifoldShap, do not.
Abstract
Machine learning pipelines commonly flatten relational data into single-table representations, discarding structural constraints. Widely used Shapley value-based feature attributions then rely on feature independence, evaluating the model on combinations that could never arise in the underlying data, producing misleading explanations. We propose RelShap, a framework that incorporates relational constraints and data provenance into Shapley value computation, restricting both background data and coalition evaluation to relationally valid configurations. The framework is estimator-agnostic and composes with Kernel SHAP, Monte Carlo, and Leverage SHAP without altering their sampling or weighting properties. Functional dependencies further induce equivalence classes over feature coalitions, which RelShap exploits to reduce runtime without changing Shapley values; we provide a combinatorial characterization of the expected speedup. Experiments across multiple datasets, models, and estimators show that RelShap produces explanations that are more faithful to the data-generating process, correctly identifying the dominant feature in controlled settings where existing methods, including Conditional SHAP and ManifoldShap, do not. Our code is available at: https://github.com/duneag2/relshap.
It is proved that sufficiently parameterized RGNNs contain sparse subnetworks that maintain 1-RWL expressivity and derive a lower bound on the probability that a random pruning yields such a subnetwork.
Lorenz Kummer, Samir Moustafa, Anatol Ehrlich et al.· 0 citations
Interpreting machine learning models fairly and efficiently remains challenging, particularly when features are correlated. Classical Shapley-based explanations can split attribution among substitutes and are often computationally demanding. This study presents iAdditive, a model-agnostic approach that promotes fairness by grouping highly correlated or redundant features through group-level attribution semantics and improves efficiency via a dynamic coalition heuristic inspired by additive explanation methods. Experiments on simulated datasets with known structure and on NHANES indicate that iAdditive produces faithful global attributions under correlation while achieving substantial runtime reductions compared with KernelSHAP, TreeSHAP, SAGE, and exact Shapley baselines. By balancing fairness, interpretability, and efficiency, iAdditive provides a practical tool for trustworthy decision support in applications such as health care.
Abirami Gunasekaran, Minsi Chen, Richard Hill et al.· International Journal of Dat...· 0 citations
This work provides a mathematically grounded, highly efficient diagnostic tool to uncover human label failures, sanitize evaluation benchmarks, and ensure the integrity of LLM alignment data.
Yunting Song, Matthew Watson, Peter Grabowski et al.· arXiv.org· 0 citations
FlowNeg is introduced, a context-conditioned hierarchical generative flow network that amortizes reward-proportional sampling without normalizing a composite reward over the entity set: given a positive triple and corruption side, it selects a type, then an entity.
Data valuation is essential for quantifying data worth in machine learning, especially in multi-party collaborations where incentives rely on fair contribution attribution. Although Shapley-value-based methods are effective for I.I.D. data, extending them to heterogeneous graphs remains challenging due to type-dependent node roles, cross-type dependencies, and prohibitive valuation costs. In this paper, we propose LTSV (Layered Type-Constrained Shapley Value, a framework that decomposes each node's value into three components: intra-type feature contribution, cross-type semantic bridging, and global structural support. Specifically, LTSV anchors the valuation in type-constrained subgames to ensure axiomatic fairness for within-type contributions while preserving the heterogeneous context. It further introduces meta-path-based value propagation to make the implicit contribution of unlabeled bridge nodes explicit and quantifiable. To scale valuation, we develop type-constrained Monte Carlo sampling with importance weighting. Experiments on DBLP, ACM, and IMDb show that LTSV more accurately identifies high-value nodes than state-of-the-art baselines under standard deletion/addition evaluations, while achieving favorable computational efficiency.
Xin Tang, Quanyan Gao, Chao Li· Proceedings of the 32nd ACM...· 0 citations
