论文检索

Jiajie Su, Chaochao Chen 0001, Weiming Liu 0005, Yuhang Wang, Yihao Wang 0007, Zheqi Lv, Xiaolin Zheng, Jianwei Yin

Session-based Recommendation (SBR) explores dynamic user interests based on short anonymous sessions. The performance of traditional ID-collaborative SBR models is constrained by the limited session length. To this end, we focus on Multi-modal Session-based Recommendation (MMSBR), which aims to leverage modality knowledge to promote short-term user interest modeling. Previous studies cannot solve the MMSBR problem well due to issues as: (I1) Modal-invariant and specific dependencies are difficult to disentangle and unify. (I2) Multiple modal transitions produce mutual effects. (I3) Modal-aware personalized preference discrepancy exists. Thus, we propose a modality-aware diffusion-based framework MDSD with subspace disentanglement, which decouples multi-modal collaborations and provides a new paradigm that integrates modality representation generation into next-item prediction. We first employ modality-disentangled consistency graphs based on affinity semantics to extract the unified modal subspace. Then we design a cross-modal contrastive attention fusion that explores cross-modality interplay, enhancing modal alignment and consistency. To interpret personalized interest discrepancy, we realize the preference-guided multi-modal diffusion, which combines modality-aware interest generation with prediction based on collaborative signals, thus providing comprehensive recommendation results. Extensive experiments on real-world datasets demonstrate the effectiveness of MDSD.

Will Stamey, Sriram Somanchi, Edward McFowland III

Difference-in-differences (DiD) has been extensively applied in the literature to elicit the average causal effect of an intervention or policy. Though researchers explore heterogeneity in the treatment effect with respect to time or some observed covariate (usually driven by domain knowledge), there is limited work on a principled, algorithmic approach to discovering the subgroups that exhibit heterogeneity in the treatment effect in DiD settings. In this research, we propose the Difference-in-Differences Subset Scan (DiD-Scan), which finds subregions of the observed covariate space corresponding to anomalous patterns in the conditional mean treatment effect. We mold our method to the DiD setting by also enabling a scan for dynamic effects over time and across multiple outcome variables. This supports the discovery of patterns where a subset of outcomes are highly impacted for a limited time window. We develop a generalized likelihood ratio-based scoring function to quantify the treatment effect of a given subgroup and propose a computationally efficient method to discover the subgroups with the largest treatment effect. We extend the method to consider correlations across time, a common condition in difference-in-difference settings that increases the difficulty of subset identification. Lastly, we demonstrate the efficacy and interpretability of the method with both simulations and applications to real datasets, replicating and extending two published studies.

Yixin Ren, Chenghou Jin, Yewei Xia, Li Ke, Longtao Huang, Hui Xue 0001, Hao Zhang 0079, Jihong Guan, Shuigeng Zhou

Determining conditional independence (CI) relationships between random variables is a fundamental yet challenging task in machine learning and statistics, especially in high-dimensional settings. Existing generative model-based CI testing methods, such as those utilizing generative adversarial networks (GANs), often struggle with undesirable modeling of conditional distributions and training instability, resulting in subpar performance. To address these issues, we propose a novel CI testing method via score-based generative modeling, which achieves precise Type I error control and strong testing power. Concretely, we first employ a sliced conditional score matching scheme to accurately estimate conditional score and use Langevin dynamics conditional sampling to generate null hypothesis samples, ensuring precise Type I error control. Then, we incorporate a goodness-of-fit stage into the method to verify generated samples and enhance interpretability in practice. We theoretically establish the error bound of conditional distributions modeled by score-based generative models and prove the validity of our CI tests. Extensive experiments on both synthetic and real-world datasets show that our method significantly outperforms existing state-of-the-art methods, providing a promising way to revitalize generative model-based CI testing.

Jiahua Rao, Hanjing Lin, Jiancong Xie, Zhen Wang 0036, Shuangjia Zheng, Yuedong Yang

Graph Neural Networks (GNNs) have gained considerable traction for modeling molecular structures and predicting properties, but their interpretability remains a significant challenge in understanding chemical behaviors. Current interpretation methods often rely on post-hoc explanations, which aim to provide transparency in GNN decisions. However, these approaches struggle with interpreting complex subgraphs and fail to leverage explanations to enhance predictive capabilities. While transparent methods can enhance GNN predictions, they typically compromise on explanation precision. This limitation underscores the need for a new strategy that effectively integrates GNN explanations and predictions. In this study, we have developed a novel interpretable causal GNN framework that combines retrieval-based causal learning with Graph Information Bottleneck (GIB) theory. Our framework semi-parametrically identifies crucial subgraphs through GIB and compresses explanatory subgraphs using a causal module. The framework consistently outperformed state-of-the-art methods, achieving a 32.72% increase in precision for scientific explanation tasks involving diverse substructures. More importantly, the learned explanations were also shown to be able to improve GNN prediction performance. This advancement is particularly vital for molecular graph learning, as it addresses the critical need to interpret how molecular structures influence predicted properties, thereby aiding drug discovery and materials science by providing insights into chemical mechanisms.

Dazhuo Qiu, Jinwen Chen, Arijit Khan 0001, Yan Zhao 0008, Francesco Bonchi

Counterfactual learning is emerging as an important paradigm, rooted in causality, which promises to alleviate common issues of graph neural networks (GNNs), such as fairness and interpretability. However, as in many real-world application domains where conducting randomized controlled trials is impractical, one has to rely on available observational (factual) data to detect counterfactuals. In this paper, we introduce and tackle the problem of searching for counterfactual evidences for the GNN-based node classification task. A counterfactual evidence is a pair of nodes such that, regardless they exhibit great similarity both in the features and in their neighborhood subgraph structures, they are classified differently by the GNN. We develop effective and efficient search algorithms and a novel indexing solution that leverages both node features and structural information to identify counterfactual evidences, and generalizes beyond any specific GNN. Through various downstream applications, we demonstrate the potential of counterfactual evidences to enhance fairness and accuracy of GNNs.

Jie Peng 0005, Zhewei Wei, Yuhang Ye 0002

Due to the proficiency of self-attention mechanisms (SAMs) in capturing dependencies in sequence modeling, several existing dynamic graph neural networks (DGNNs) utilize Transformer architectures with various encoding designs to capture sequential evolutions of dynamic graphs. However, the effectiveness and efficiency of these Transformer-based DGNNs vary significantly, highlighting the importance of properly defining the SAM on dynamic graphs and comprehensively encoding temporal and interactive dynamics without extra complex modules. In this work, we propose TIDFormer, a dynamic graph TransFormer that fully exploits Temporal and Interactive Dynamics in an efficient manner. We clarify and verify the interpretability of our proposed SAM, addressing the open problem of its uninterpretable definitions on dynamic graphs in previous works. To model the temporal and interactive dynamics, respectively, we utilize the calendar-based time partitioning information and extract informative interaction embeddings for both bipartite and non-bipartite graphs using merely the sampled first-order neighbors. In addition, we jointly model temporal and interactive features by capturing potential changes in historical interaction patterns through a simple decomposition. We conduct extensive experiments on several dynamic graph datasets to verify the effectiveness and efficiency of TIDFormer. The experimental results demonstrate that TIDFormer excels, outperforming state-of-the-art models across most datasets and experimental settings. Furthermore, TIDFormer exhibits significant efficiency advantages compared to previous Transformer-based methods.

Yong-chan Park, Kisoo Kim, U Kang

How can we achieve compact tensor representations without sacrificing reconstruction accuracy? Tensor decomposition is a cornerstone of modern data mining and machine learning, enabling efficient representations of multi-dimensional data through fundamental algorithms such as CP, Tucker, and Tensor-Train decompositions. However, directly applying these methods to raw data often results in high target ranks, poor reconstruction accuracy, and computational inefficiencies, as the data may not naturally conform to the low-rank structures these methods assume. In this paper, we propose PuzzleTensor, a method-agnostic data transformation technique for compact tensor factorization. Given a data tensor, PuzzleTensor ''solves the puzzle'' by shifting each hyperslice of the tensor to achieve accurate decompositions with significantly lower target ranks. PuzzleTensor offers three key advantages: (1) it is independent of specific decomposition methods, making it seamlessly compatible with various algorithms, such as CP, Tucker, and Tensor-Train decompositions; (2) it works under weak data assumptions, showing robust performance across both sparse and dense data, regardless of the rank; (3) it is inherently explainable, allowing clear interpretation of its learnable parameters and layer-wise operations. Extensive experiments show that PuzzleTensor consistently outperforms direct tensor decomposition approaches by achieving lower reconstruction errors and reducing the required target rank, making it a versatile and practical tool for compact tensor factorization in real-world applications.

Jiri Nemecek 0002, Mark Kozdoba, Illia Kryvoviaz, Tomás Pevný, Jakub Marecek

Bias evaluation is fundamental to trustworthy AI, both in terms of checking data quality and in terms of checking the outputs of AI systems. In testing data quality, for example, one may study the distance of a given dataset, viewed as a distribution, to a given ground-truth reference dataset. However, classical metrics, such as the Total Variation and the Wasserstein distances, are known to have high sample complexities and, therefore, may fail to provide a meaningful distinction in many practical scenarios. In this paper, we propose a new notion of distance, the Maximum Subgroup Discrepancy (MSD). In this metric, two distributions are close if, roughly, discrepancies are low for all feature subgroups. While the number of subgroups may be exponential, we show that the sample complexity is linear in the number of features, thus making it feasible for practical applications. Moreover, we provide a practical algorithm for evaluating the distance based on Mixed-integer optimization (MIO). We also note that the proposed distance is easily interpretable, thus providing clearer paths to fixing the biases once they have been identified. Finally, we describe a natural general bias detection framework, termed MSDD distances, and show that MSD aligns well with this framework. We empirically evaluate MSD by comparing it with other metrics and by demonstrating the above properties of MSD on real-world datasets.

Yingtao Luo, Shikai Fang, Binqing Wu, Qingsong Wen, Liang Sun 0001

Weather forecasting is essential but remains computationally intensive and physically incomplete in traditional numerical weather prediction (NWP) methods. Deep learning (DL) models offer efficiency and accuracy but often ignore physical laws, limiting interpretability and generalization. We propose PhyDL-NWP, a physics-guided deep learning framework that integrates physical equations with latent force parameterization into data-driven models. It predicts weather variables from arbitrary spatiotemporal coordinates, computes physical terms via automatic differentiation, and uses a physics-informed loss to align predictions with governing dynamics. PhyDL-NWP enables resolution-free downscaling by modeling weather as a continuous function and fine-tunes pre-trained models with minimal overhead, achieving up to 170× faster inference with only 55K parameters. Experiments show that PhyDL-NWP improves both forecasting performance and physical consistency.

Yu Liu 0016, Weiyao Tao, Tong Xia, Simon Knight 0005, Tingting Zhu 0001

Survival analysis, which estimates the probability of event occurrence over time from censored data, is fundamental in numerous real-world applications, particularly in high-stakes domains such as healthcare and risk assessment. Despite advances in numerous survival models, quantifying the uncertainty of predictions from these models remains underexplored and challenging. The lack of reliable uncertainty quantification limits the interpretability and trustworthiness of survival models, hindering their adoption in clinical decision-making and other sensitive applications. To bridge this gap, in this work, we introduce SurvUnc, a novel meta-model based framework for post-hoc uncertainty quantification for survival models. SurvUnc introduces an anchor-based learning strategy that integrates concordance knowledge into meta-model optimization, leveraging pairwise ranking performance to estimate uncertainty effectively. Notably, our framework is model-agnostic, ensuring compatibility with any survival model without requiring modifications to its architecture or access to its internal parameters. Especially, we design a comprehensive evaluation pipeline tailored to this critical yet overlooked problem. Through extensive experiments on four publicly available benchmarking datasets and five representative survival models, we demonstrate the superiority of SurvUnc across multiple evaluation scenarios, including selective prediction, misprediction detection, and out-of-domain detection. Our results highlight the effectiveness of SurvUnc in enhancing model interpretability and reliability, paving the way for more trustworthy survival predictions in real-world

David Liu 0006, Arjun Seshadri, Tina Eliassi-Rad, Johan Ugander

A wide range of graph embedding objectives decompose into two components: one that enforces similarity, attracting the embeddings of nodes that are perceived as similar, and another that enforces dissimilarity, repelling the embeddings of nodes that are perceived as dissimilar. Without repulsion, the embeddings would collapse into trivial solutions. Skip-Gram Negative Sampling (SGNS) is a popular and efficient repulsion approach that prevents collapse by repelling each node from a sample of dissimilar nodes. In this work, we show that when repulsion is most needed and the embeddings approach collapse, SGNS node-wise repulsion is, in the aggregate, an approximate re-centering of the node embedding dimensions. Such dimension operations are more scalable than node operations and produce a simpler geometric interpretation of the repulsion. Our theoretical result establishes dimension regularization as an effective and more efficient, compared to skip-gram node contrast, approach to enforcing dissimilarity among embeddings of nodes. We use this result to propose a flexible algorithm augmentation framework that improves the scalability of any existing algorithm using SGNS. The framework prioritizes node attraction and replaces SGNS with dimension regularization. We instantiate this generic framework for LINE and node2vec and show that the augmented algorithms preserve downstream link-prediction performance while reducing GPU memory usage by up to 33.3% and training time by 23.4%. Moreover, we show that completely removing repulsion (a special case of our augmentation framework) in LINE reduces training time by 70.9% on average, while increasing link prediction performance, especially for graphs that are globally sparse but locally dense. Global sparsity slows down dimensional collapse, while local density ensures that node attraction brings the nodes near their neighbors. In general, however, repulsion is needed, and dimension regularization provides an efficient alternative to SGNS.

Brian Liu 0002, Rahul Mazumder

We present MOSS, a multi-objective optimization framework for constructing stable sets of decision rules. MOSS incorporates three important criteria for interpretability: sparsity, accuracy, and stability, into a single multi-objective optimization framework. Importantly, MOSS allows a practitioner to rapidly evaluate the trade-off between accuracy and stability in sparse rule sets in order to select an appropriate model. We develop a specialized cutting plane algorithm in our framework to rapidly compute the Pareto frontier between these two objectives, and our algorithm scales to problem instances beyond the capabilities of commercial optimization solvers. Our experiments show that MOSS outperforms state-of-the-art rule ensembles in terms of both predictive performance and stability.

Yilin Li 0018, Lu Deng, Yong Wang

Interference in online controlled experiments poses significant challenges, as treatment effects may propagate through user interactions. This phenomenon systematically biases conventional treatment effect estimators and substantially complicates the interpretation of experimental results. Detecting the presence of interference is crucial for ensuring the reliability of conclusions, ultimately enhancing decision-making processes. We formalize a framework for detecting the existence of interference with dyadic data, which is ubiquitous in social network platforms. Dyadic data, typically generated from user interactions, have not been fully utilized in online controlled experiments. We first introduce the dyadic-robust inference techniques, then illustrate how the dyadic outcomes link to the user-level metrics and how interference manifests. We then propose a comprehensive approach to detect interference effects based on dyadic-robust t-statistics, and we also provide a subsampling approach to reduce the computational complexity of the proposed methods. We also provide several approaches to leverage pre-experiment data, such as selecting the optimal treatment probability, designing stratified randomization in the pre-experiment stage, or reducing the variance with both user-level and dyad-level control variates at the post-experiment stage. Our approach is computationally efficient and easily integrated with existing pipelines. We provide theoretical guarantees for our inference procedure across various asymptotic regimes. Finally, we conduct several empirical evaluations under various dyadic data-generating processes to assess the performance and robustness of the proposed procedures. Finally, we introduce an interference detection pipeline and demonstrate it with an experiment from Weixin.

Jinning Li 0001, Ruipeng Han, Jingying Zeng, Dachun Sun, Chenkai Sun, Hanghang Tong, ChengXiang Zhai, Boleslaw K. Szymanski, Tarek F. Abdelzaher

The perceptions and decisions of individuals on social networks are deeply rooted in their intrinsic beliefs, which makes it possible to infer social beliefs from user behavior and message interactions. While existing research models these interactions as graphs and learns their representations, interpretability remains a significant challenge. In real-world scenarios, the interpretation of beliefs is nested within subject scopes of different granularity (such as topics and locations), posing additional challenges for belief discovery. In this paper, we introduce the Interpretable Graph Auto-Encoder Tree (IGAT), a novel end-to-end framework that jointly encodes hierarchical subject scopes and corresponding beliefs as a unified, interpretable hierarchical representation. IGAT integrates the interpretable hierarchy of Model Trees with disentangled representation learning models. We propose a differentiable Slice Mechanism to dynamically optimize internal node splitting and jointly train a leaf model to learn disentangled belief subspaces. The aggregation of these subspaces yields a unified representation, offering interpretations for both subjects and beliefs. Experimental evaluations on three real-world Twitter datasets show that IGAT achieves a consistent improvement of 1.49%-5.61% in F1-score, accuracy, and purity in the belief discovery task, as well as its effectiveness in various downstream analytical applications.

Bingheng Li, Zhikai Chen, Haoyu Han 0001, Shenglai Zeng, Jingzhe Liu, Jiliang Tang

A fundamental challenge in understanding graph neural networks (GNNs) lies in characterizing their optimization dynamics and loss landscape geometry, critical for improving interpretability and robustness. While mode connectivity-a lens for analyzing geometric properties of loss landscapes-has proven insightful for other deep learning architectures, its implications for GNNs remain unexplored. This work presents the first investigation of mode connectivity in GNNs. We uncover that GNNs exhibit distinct non-linear mode connectivity, diverging from patterns observed in fully-connected networks or CNNs. Crucially, we demonstrate that graph structure, rather than model architecture, dominates this behavior, with graph properties like homophily correlating with mode connectivity patterns. We further establish a link between mode connectivity and generalization, proposing a generalization bound based on loss barriers and revealing its utility as a diagnostic tool. Our findings further bridge theoretical insights with practical implications: they rationalize domain alignment strategies in graph learning and provide a foundation for refining GNN training paradigms.

Hector Kohler, Riad Akrour, Philippe Preux

In supervised learning, decision trees are valued for their interpretability and performance. While greedy decision tree algorithms like CART remain widely used due to their computational efficiency, they often produce sub-optimal solutions with respect to a regularized training loss. Conversely, optimal decision tree methods can find better solutions but are computationally intensive and typically limited to shallow trees or binary features. We present Dynamic Programming Decision Trees (DPDT), a framework that bridges the gap between greedy and optimal approaches. DPDT relies on a Markov Decision Process formulation combined with heuristic split generation to construct near-optimal decision trees with significantly reduced computational complexity. Our approach dynamically limits the set of admissible splits at each node while directly optimizing the tree regularized training loss. Theoretical analysis demonstrates that DPDT can minimize regularized training losses at least as well as CART. Our empirical study shows on multiple datasets that DPDT achieves near-optimal loss with orders of magnitude fewer operations than existing optimal solvers. More importantly, extensive benchmarking suggests statistically significant improvements of DPDT over both CART and optimal decision trees in terms of generalization to unseen data. We demonstrate DPDT practicality through applications to boosting, where it consistently outperforms baselines. Our framework provides a promising direction for developing efficient, near-optimal decision tree algorithms that scale to practical applications.

Ricardo Knauer, Mario Koddenbrock, Raphael Wallsberger, Nicholas M. Brisson, Georg N. Duda, Deborah Falla, David W. Evans, Erik Rodner

Large language models (LLMs) provide powerful means to leverage prior knowledge for predictive modeling when data is limited. In this work, we demonstrate how LLMs can use their compressed world knowledge to generate intrinsically interpretable machine learning models, i.e., decision trees, without any training data. We find that these zero-shot decision trees can even surpass data-driven trees on some small-sized tabular datasets and that embeddings derived from these trees perform better than data-driven tree-based embeddings on average. Our decision tree induction and embedding approaches can therefore serve as new knowledge-driven baselines for data-driven machine learning methods in the low-data regime. Furthermore, they offer ways to harness the rich world knowledge within LLMs for tabular machine learning tasks. Our code and results are available at https://github.com/ml-lab-htw/llm-trees.

Makoto Imamura

Time series anomaly detection (TSAD) is a central topic in data mining, with many algorithms proposed, including deep learning and LLM-based methods. However, these are often evaluated on flawed benchmarks with trivial anomalies, unrealistic densities, mislabeled ground truth, and run-to-failure biases. Recent studies using higher-quality datasets show that simpler distance-based methods, such as discords, can outperform deep learning; However, they typically rely on sliding windows with fixed lengths, limiting flexibility. We introduce GDFlex (Generalized Discords with Flexible Subsequence Length), a discord-based method that detects anomalies across multiple time scales using length-normalized Euclidean distance. It also incorporates bias-correction mechanisms to address limitations in Euclidean distance and z-normalization. GDFlex achieves notable accuracy gains over existing methods while remaining scalable, interpretable, and efficient.

Shaoyuan Huang, Tiancheng Zhang 0009, Zhongtian Zhang, Xiaofei Wang 0001, Lanjun Wang, Xin Wang 0030

Time series forecasting is a critical and practical problem in many real-world applications, especially for industrial scenarios, where load forecasting underpins the intelligent operation of modern systems like clouds, power grids and traffic networks. However, the inherent complexity and dynamics of these systems present significant challenges. Despite advances in methods such as pattern recognition and anti-non-stationarity have led to performance gains, current methods fail to consistently ensure effectiveness across various system scenarios due to the intertwined issues of complex patterns, concept-drift, and few-shot problems. To address these challenges simultaneously, we introduce a novel scheme centered on fundamental waveform, a.k.a., meta-pattern. Specifically, we develop a unique Meta-pattern Pooling mechanism to purify and maintain meta-patterns, capturing the nuanced nature of system loads. Complementing this, the proposed Echo mechanism adaptively leverages the meta-patterns, enabling a flexible and precise pattern reconstruction. Our Meta-pattern Echo transformer (MetaEformer) seamlessly incorporates these mechanisms with the transformer-based predictor, offering end-to-end efficiency and interpretability of core processes. Demonstrating superior performance across eight benchmarks under three system scenarios, MetaEformer marks a significant advantage in accuracy, with a 37% relative improvement on fifteen state-of-the-art baselines.

Yile Chen 0001, Yicheng Tao 0001, Yue Jiang 0005, Shuai Liu 0018, Han Yu 0001, Gao Cong

The widespread adoption of location-based services has led to the generation of vast amounts of mobility data, providing significant opportunities to model user movement dynamics within urban environments. Recent advancements have focused on adapting Large Language Models (LLMs) for mobility analytics. However, existing methods face two primary limitations: inadequate semantic representation of locations (i.e., discrete IDs) and insufficient modeling of mobility signals within LLMs (i.e., single templated instruction fine-tuning). To address these issues, we propose QT-Mob, a novel framework that significantly enhances LLMs for mobility analytics. QT-Mob introduces a location tokenization module that learns compact, semantically rich tokens to represent locations, preserving contextual information while ensuring compatibility with LLMs. Furthermore, QT-Mob incorporates a series of complementary fine-tuning objectives that align the learned tokens with the internal representations in LLMs, improving the model's comprehension of sequential movement patterns and location semantics. The proposed QT-Mob framework not only enhances LLMs' ability to interpret mobility data but also provides a more generalizable approach for various mobility analytics tasks. Experiments on three real-world dataset demonstrate the superior performance in both next-location prediction and mobility recovery tasks, outperforming existing deep learning and LLM-based methods.