Large language models (LLMs) have demonstrated impressive capabilities in utilizing external tools.In practice, however, LLMs are often exposed to tools that are irrelevant to the user's query, in which case the desired behavior is to refrain from invocations.In this work, we identify a widespread yet overlooked mechanistic flaw in tool refusal, which we term structural alignment bias: Even when a tool fails to serve the user's goal, LLMs still tend to invoke it whenever query attributes can be validly assigned to tool parameters.To systematically study this bias, we introduce SABEval, a new dataset that decouples structural alignment from semantic relevance.Our analysis shows that structural alignment bias induces severe tool-invocation errors in LLMs, yet remains largely unaccounted for in existing evaluations.To investigate the internal mechanisms underlying this bias, we propose Contrastive Attention Attribution, which reveals two competing pathways for semantic checking and structural matching.The relative strength of these pathways drives LLMs' tool invocation decisions.Based on these findings, we further introduce a rebalancing strategy that effectively mitigates structural alignment bias, as demonstrated by extensive experiments, without degrading general tool-use capabilities.Our code and dataset are available at https://github. com/along-l/irrelevant-tool. Can you tell me if 'Spider-Man: Miles Morales' is on sale on the Canadian PlayStation Store?"name": "check_nintendo_switch_game_on_sale", "description": "Checks if a particular game is currently on sale on the Nintendo Switch Store.","parameters": { "game_title": {"type": "string", "description": "The title of the video game."},"region": {"type": "string", "description": "The region to check for the Nintendo Switch Store."}}

Entity alignment (EA) aims to identify entities across different knowledge graphs (KGs) that refer to the same real-world object and plays a critical role in knowledge fusion and integration.Traditional EA methods mainly rely on knowledge representation learning, but their performance is often limited under noisy or sparsely supervised scenarios.Recently, large language models (LLMs) have been introduced to EA and achieved notable improvements by leveraging rich semantic knowledge.However, existing LLM-based EA approaches typically treat LLMs as black-box decision makers, resulting in limited interpretability, and the direct use of large-scale triples substantially increases inference cost.To address these challenges, we propose EA-Agent, a reasoningdriven agent for EA.EA-Agent formulates EA as a structured reasoning process with multistep planning and execution, enabling interpretable alignment decisions.Within this process, it introduces attribute and relation triple selectors to filter redundant triples before feeding them into the LLM, effectively addressing efficiency challenges.Experimental results on three benchmark datasets demonstrate that EA-Agent consistently outperforms existing EA methods and achieves state-of-the-art performance.The source code is available at https: //github.com/YXNan0110/EA-Agent.

Single-expert fake news detectors, such as Graph Neural Networks (GNNs) and Large Language Models (LLMs), increasingly struggle to counter the diversifying camouflage tactics of modern adversaries, which range from semantic (e.g., mimicking writing styles) to structural (e.g., manipulating propagation paths). To address this, existing methods attempt to build a hybrid model by sequentially incorporating GNNs and LLMs; however, such hybridization blurs the distinction between experts and prevents critical cross-validation. In addition, existing methods rely heavily on vast labeled data, which is costly to acquire, particularly for fake news samples. In this paper, we propose a Semi-supervised Mixture of Experts framework for Fake news detection, namely S2MOE-F. The core idea of S2MOE-F is to establish a robust defense against multifaceted camouflage by cross-validating the complementary judgments of two independent experts, GNN and LLM. On the one hand, S2MOE-F drives experts' judgments by using a One-Class Classification (OCC) objective, which constrains true news within a compact hypersphere and identifies samples outside this boundary as fake, reducing reliance on scarce fake news labels. On the other hand, S2MOE-F generates high-confidence pseudo-labels based on consensus or divergence between experts to exploit abundant unlabeled data. In addition, we propose a novel reinforcement learning (RL)-based routing policy that dynamically determines the dominant expert for input samples without explicit supervision. Finally, we design a disentangled masked Transformer to ensure experts' specialization by reducing inter-expert redundancy. Extensive experiments on real-world datasets sourced from Web platforms and social media demonstrate the superior performance of S2MOE-F.

Graph-level anomaly detection (GLAD) is a critical task to identify graphs with abnormal properties in various domains, ranging from fraudulent social networks to malicious botnets on online platforms. The dominant paradigm for existing GLAD detectors has been partially unsupervised, relying on training data composed exclusively of normal samples. However, this partially unsupervised paradigm inevitably requires a costly expert filtering process to ensure the training data is free of anomalies. This creates a significant gap between current approaches and the real-world necessity of a fully unsupervised paradigm, which involves training a model directly on real-world data ''as-is'', with its inherent mix of normal and anomalous samples. To bridge this gap, we incorporate uncertainty learning into GLAD to promote fully unsupervised learning. We propose two frameworks: Score Uncertainty Learning (SUL) and Graph-data Uncertainty Learning (GUL). Specifically, SUL enhances existing GLAD detectors by modeling uncertainty through Gaussian distributions over the detectors' predictions, adaptively attenuating the influence of potential anomalies. GUL is an end-to-end framework that iteratively optimizes anomaly detection and uncertainty modeling via an Expectation-Maximization algorithm. In addition, we develop a dedicated loss that utilizes potential anomalies to enhance the effectiveness and robustness of GUL. Empirical results on sixteen benchmark datasets, covering real-world graphs from social networks and online platforms, demonstrate the superiority of our methods and highlight the promise of incorporating uncertainty into fully unsupervised GLAD.

Graph-level anomaly detection (GLAD) aims to identify graphs that significantly deviate from others in a graph dataset. Existing methods predominantly rely on standard Graph Neural Networks (GNNs) to learn graph representations, but they often overlook subgraph-level information, which provides essential structural and semantic cues for distinguishing normal and anomalous graphs. This limitation not only compromises the detection performance but also hinders the interpretability of GLAD predictions. To address these challenges, we propose NGLAD, a novel framework that introduces the concept of normality-relevant subgraphs that capture shared patterns across normal graphs. These subgraphs serve as key indicators to distinguish normal graphs from anomalies that that often lack or deviate from such patterns. During model training, by explicitly modeling the shared subgraph patterns inherent in normal graphs through a Subgraph Extractor and a Normality Learner, NGLAD identifies the subgraphs most relevant to normality. Leveraging the One-class Information Bottleneck principle, these modules ensure that the extracted subgraphs retain only the most informative features of normality while filtering out irrelevant nodes and edges. During inference, NGLAD detects anomalies by evaluating inconsistencies in representations between the input graph and its extracted subgraph. Extensive evaluations on synthetic and real-world datasets demonstrate that NGLAD significantly outperforms state-of-the-art methods in detection performance while offering interpretable explanations.

Graph-level anomaly detection (GLAD) identifies graphs exhibiting abnormal properties within a graph dataset.Despite promising results in this task, the state-of-the-art methods cannot be fully trusted and deployed in realistic scenarios due to their black-box nature.To alleviate this, existing methods try to explain predictions by extracting important subgraphs from each graph, as instancelevel explanations.However, instance-level explanations across all samples are costly to verify and insufficient to capture the model's general behaviors.Thus, we propose a global interpretable Graph-Level Anomaly Detection model via Prototype (GLADPro), which provides global-level explanations throughout the entire dataset, that is, the significant subgraph patterns that consistently influence the model's decisions.Specifically, GLADPro incorporates prototype learning with the information bottleneck principle, enabling prototypes to capture the most significant subgraph patterns as global-level explanations through persistent interactions with key subgraphs from input graphs.In addition, a regularization term is proposed to prevent the collapse traps with theoretical proof.Finally, we filter redundant prototypes using the maximum mean discrepancy metric.Extensive experiments demonstrate the superiority of GLADPro in anomaly detection and explainability; for instance, on the mutagen dataset, it reduces the number of explanations to verify from 1403 to only 6.

Graph-level out-of-distribution (OOD) detection, which attempts to identify OOD graphs originated from an unknown distribution, is a vital building block for safety-critical applications in Web and society. Current approaches concentrate on how to learn better graph representations, but fail to provide any statistically guarantee on detection results, therefore impeding their deployments in the scenario where detection errors would result in serious consequences. To overcome this critical issue, we propose the Conformal Graph-level Out-of-distribution Detection (CGOD), extending the theory of conformal prediction to graph-level OOD detection with a rigorous control over the false positive rate. In CGOD, we develop a new aggregated non-conformity score function based on the proposed adaptive data augmentation. Through the guidance from two designed metrics, i.e., score consistency and representation diversity, our augmentation strategy can generate multiple non-conformity scores, and aggregating these generated non-conformity scores together is robust to the misleading information. Meanwhile, our score function can perceive the subsequent process of conformal inference, enabling the aggregated non-conformity score to be adaptive to different input graphs and deriving a more accurate score estimation. We conduct experiments on multiple real-world datasets with different empirical settings. Extensive results and model analyses demonstrate the superior performance of our approach over several competitive baselines.

Graph neural networks (GNNs) have shown promise in graph classification tasks, but they struggle to identify out-of-distribution (OOD) graphs often encountered in real-world scenarios, posing a significant obstacle for their open-world deployment. Due to the unpredictable nature of the various distributions to which OOD graphs adhere, the challenge of OOD graph detection lies in enabling models to capture distribution differences between in-distribution (ID) and OOD graphs. Current methods often introduce a subset of OOD patterns, such as synthetic OOD graphs, to facilitate learning the discrimination between ID and OOD graphs. However, these OOD patterns may not sufficiently encapsulate the entire range of OOD graphs, leading to inadequate learning of the distribution differences between ID and OOD graphs. In this article, we propose a novel OOD graph detection algorithm, ODGNN. The ODGNN does not expose GNNs to any OOD patterns during model training, thus reducing bias toward specific types of OOD graph samples and enhancing OOD graph detection. The algorithm differentiates graphs by evaluating whether the input graphs conform to established ID graph class-conditioned distributions. Specifically, during model training, the ODGNN integrates a Gaussian encoder into GNNs to characterize ID graph classes using distinct class-conditioned distributions. During inference, OOD graphs are mapped to a representation space distant from ID graphs due to their divergence from any known class-conditioned distribution. Extensive experiments conducted on real-world datasets validate the effectiveness of the ODGNN in enhancing OOD detection performance across various GNN-based graph classification models. The ODGNN also demonstrates superior performance compared to state-of-the-art OOD graph detection competitors.

Network alignment involves identifying corresponding nodes across multiple networks. The majority of existing methods adhere to the assumption of consistency. However, due to distinct graph generation mechanisms, anchor nodes in real-world datasets often exhibit more intricate structural patterns, such as having multiple different neighbors and higher-order associations. Relying solely on consistency while disregarding the intricate patterns of anchor links may potentially inflict substantial detriment upon both the accuracy of network alignment and the generality of the model. In this paper, we introduce the disparity and diversity based on distinct structural patterns of ubiquitous anchor links. We propose a comprehensive framework that employs first-order proximity, lower-order discriminability, and higher-order correlation to model consistency, disparity, and diversity. We also incorporate a post-fusion mechanism for effectively integrating alignment matrices. Furthermore, we innovatively introduce hyperbolic space as an embedding space to further minimize embedding distortion. Extensive experiments have shown that our approach achieves state-of-the-art alignment results and yields notable improvements in the overall versatility of the model.

Graph neural networks (GNNs) are now the mainstream method for mining graph-structured data and learning low-dimensional node- and graph-level embeddings to serve downstream tasks. However, limited by the bottleneck of interpretability that deep neural networks present, existing GNNs have ignored the issue of estimating the appropriate number of dimensions for the embeddings. Hence, we propose a novel framework called Minimum Graph Entropy principle-guided Dimension Estimation, i.e. MGEDE, that learns the appropriate embedding dimensions for both node and graph representations. In terms of node-level estimation, a minimum entropy function that counts both structure and attribute entropy, appraises the appropriate number of dimensions. In terms of graph-level estimation, each graph is assigned a customized embedding dimension from a candidate set based on the number of dimensions estimated for the node-level embeddings. Comprehensive experiments with node and graph classification tasks and nine benchmark datasets verify the effectiveness and generalizability of MGEDE.

Graph-level anomaly detection aims to distinguish anomalous graphs in a graph dataset from normal graphs. Anomalous graphs represent a very few but essential patterns in the real world. The anomalous property of a graph may be referable to its anomalous attributes of particular nodes and anomalous substructures that refer to a subset of nodes and edges in the graph. In addition, due to the imbalance nature of anomaly problem, anomalous information will be diluted by normal graphs with overwhelming quantities. Various anomaly notions in the attributes and/or substructures and the imbalance nature together make detecting anomalous graphs a non-trivial task. In this paper, we propose a graph neural network for graph-level anomaly detection, namely iGAD. Specifically, an anomalous graph attribute-aware graph convolution and an anomalous graph substructure-aware deep Random Walk Kernel (deep RWK) are welded into a graph neural network to achieve the dual-discriminative ability on anomalous attributes and substructures. Deep RWK in iGAD makes up for the deficiency of graph convolution in distinguishing structural information caused by the simple neighborhood aggregation mechanism. Further, we propose a Point Mutual Information (PMI)-based loss function to target the problems caused by imbalance distributions. PMI-based loss function enables iGAD to capture essential correlation between input graphs and their anomalous/normal properties. We evaluate iGAD on four real-world graph datasets. Extensive experiments demonstrate the superiority of iGAD on the graph-level anomaly detection task.

With the development of e-commerce, fraud behaviors have been becoming one of the biggest threats to the e-commerce business. Fraud behaviors seriously damage the ranking system of e-commerce platforms and adversely influence the shopping experience of users. It is of great practical value to detect fraud behaviors on e-commerce platforms. However, the task is non-trivial, since the adversarial action taken by fraudsters. Existing fraud detection systems used in the e-commerce industry easily suffer from performance decay and can not adapt to the upgrade of fraud patterns, as they take already known fraud behaviors as supervision information to detect other suspicious behaviors. In this article, we propose a competitive graph neural networks (CGNN)-based fraud detection system (eFraudCom) to detect fraud behaviors at one of the largest e-commerce platforms, “Taobao” 1 . In the eFraudCom system, (1) the competitive graph neural networks (CGNN) as the core part of eFraudCom can classify behaviors of users directly by modeling the distributions of normal and fraud behaviors separately; (2) some normal behaviors will be utilized as weak supervision information to guide the CGNN to build the profile for normal behaviors that are more stable than fraud behaviors. The algorithm dependency on fraud behaviors will be eliminated, which enables eFraudCom to detect fraud behaviors in presence of the new fraud patterns; (3) the mutual information regularization term can maximize the separability between normal and fraud behaviors to further improve CGNN. eFraudCom is implemented into a prototype system and the performance of the system is evaluated by extensive experiments. The experiments on two Taobao and two public datasets demonstrate that the proposed deep framework CGNN is superior to other baselines in detecting fraud behaviors. A case study on Taobao datasets verifies that CGNN is still robust when the fraud patterns have been upgraded.

The objective of fraud detection is to distinguish fraudsters from normal users. In graph/network environments, both fraudsters and normal users are modeled as nodes, and the connections between those nodes are represented as edges. Fraudsters typically try to camouflage themselves with “normal” behaviors, say, by deliberately establishing many connections to normal users. Such camouflage inherently makes their appearance inconsistent with the essence of what it is to be normal, and gives rise to inconsistencies in the graph. In this paper, we investigate three aspects of these graph inconsistencies: features, topologies, and relations. To date, graph-based fraud detectors have shown a rather limited capability to fuse information about different types of inconsistencies. Apart from that, there is another problem of imbalance to overcome. This is because fraudsters usually only account for a very small percentage of all users. To achieve a promising capability, i.e., dual-resistant to graph inconsistency and imbalance, we present a new fraud detection model FRAUDRE based on Graph Neural Networks. Extensive experiments comparing eight up-to-date baselines on two real-world datasets, Amazon and YelpChi, demonstrate the superiority of FRAUDRE.

Finding semantic communities using network topology and contents together is a hot topic in community detection. Existing methods often use word attributes in an indiscriminate way to help finding communities. Through analysis we find that, words in networked contents often embody a hierarchical semantic structure. Some words reflect a background topic of the whole network with all communities, some imply the high-level general topic covering several topic-related communities, and some imply the high-resolution specialized topic to describe each community. Ignoring such semantic structures often leads to defects in depicting networked contents where deep semantics are not fully utilized. To solve this problem, we propose a new Bayesian probabilistic model. By distinguishing words from either a background topic or some two-level topics (i.e., general and specialized topics), this model not only better utilizes the networked contents to help finding communities, but also provides a clearer multiplex semantic community interpretation. We then give an efficient variational algorithm for model inference. The superiority of this new approach is demonstrated by comparing with ten state-of-the-art methods on nine real networks and an artificial benchmark. A case study is further provided to show its strong ability in deep semantic interpretation of communities.

Using network topology and semantic contents to find topic-related communities is a new trend in the field of community detection. By analyzing texts in social networks, we find that topics in networked contents are often hierarchical. In most cases, they have a two-level semantic structure with general and specialized topics, to respectively denote common and specific interests of communities. However, the existing community detection methods ignore such a hierarchy and take all words used to describe node semantics from an identical perspective. This indiscriminate use of words leads to natural defects in depicting networked content in which the deep semantics is not fully utilized. To address this problem, we propose a novel probabilistic generative model. By distinguishing the general and specialized topics of words, our model not only can find community structures more accurately, but also provide two-level semantic interpretation for each community. We train the model by deriving an efficient inference method under the framework of variational expectation-maximization. We provide a case study to show the ability of our algorithm in deep semantic interpretability of communities. The superiority of our algorithm for community detection is further demonstrated in comparison with eight state-of-the-art algorithms on eight real-world networks.