Social bot detection is pivotal for safeguarding the integrity of online information ecosystems. Although recent graph neural network (GNN) solutions achieve strong results, they remain hindered by two practical challenges: (i) severe class imbalance arising from the high cost of generating bots, and (ii) topological noise introduced by bots that skillfully mimic human behavior and forge deceptive links. We propose the Reinforcement-guided graph Augmentation social Bot detector (RABot), a multi-granularity graph-augmentation framework that addresses both issues in a unified manner. RABot employs a neighborhood-aware oversampling strategy that linearly interpolates minority-class embeddings within local subgraphs, thereby stabilizing the decision boundary under low-resource regimes. Concurrently, a reinforcement-learning-driven edge-filtering module combines similarity-based edge features with adaptive threshold optimization to excise spurious interactions during message passing, yielding a cleaner topology. Extensive experiments on three real-world benchmarks and four GNN backbones demonstrate that RABot consistently surpasses state-of-the-art baselines. In addition, since its augmentation and filtering modules are orthogonal to the underlying architecture, RABot can be seamlessly integrated into existing GNN pipelines to boost performance with minimal overhead.

With the rise of social media, accurately predicting information diffusion has become crucial for a wide range of applications. Existing methods usually employ sequential hypergraphs to model users' latent interaction preferences and use self-attention mechanisms to capture dependencies with users. However, they typically focus on a single temporal scale and lack the ability to effectively model temporal influence, which limits their performance in diffusion prediction tasks. To address these limitations, we propose a novel method (MHPS) to enhance information diffusion prediction via multiple granularity hypergraphs and a position-aware sequence model. Specifically, MHPS constructs hypergraph sequences of different granularities by grouping user interactions according to various time intervals. Additionally, to further enhance the modeling of temporal influence, two types of cross-attention mechanisms, namely next-step positional cross-attention and source influence cross-attention, are introduced within the cascade representation. The next-step positional cross-attention captures target position awareness, while the source influence cross-attention focuses on the impact of the initial source. Then, gating mechanisms and GRUs are employed to fuse the different attention outputs and predict the next target user. Extensive experiments on real-world datasets demonstrate that MHPS achieves competitive performance against state-of-the-art methods. The average improvements are up to 7.82% in terms of Hits@10 and 5.60% in terms of MAP@100. Our code is available at https://github.com/cgao-comp/MHPS.

With the rapid global growth of short video platforms, the spread of fake news has become increasingly prevalent, creating an urgent demand for effective automated detection methods. Current approaches typically rely on feature extractors to gather information from multiple modalities and then generate predictions through classifiers. However, these methods often fail to fully utilize the complex information across all modalities and overlook the potential for video manipulation, limiting their overall performance. To tackle these issues, MFAE is proposed, a novel framework for Multimodal Feature Adaptive Enhancement for Fake News Video Detection. The framework starts by extracting semantic and emotional features from the news, which are the basis for generating coarse multimodal representations. These representations are further refined through Adaptive Enhancement, a module specifically designed to strengthen the visual and audio modalities. Subsequently, spatial and temporal features are extracted separately, with temporal features undergoing additional refinement via a Temporal Enhancement module. The final result is obtained by feeding the individually enhanced features into the multimodal feature integration module for interaction Comprehensive experiments on two benchmark datasets highlight the exceptional performance of MFAE in detecting fake news on short video platforms. Specifically, the method achieves accuracy improvements of 2.21% and 4.35% on FakeSV and FakeTT, respectively.

Malicious social bot detection is vital for social network security. While graph neural networks (GNNs) based methods have improved performance by modeling structural information, they often overlook latent community structures, resulting in homogeneous node representations. Leveraging community structures, which capture discriminative group-level patterns, is therefore essential for more robust detection. In this paper, we propose a new Community-Aware Graph Contrastive Learning (CAGCL) framework for enhanced social bot detection. Specifically, CAGCL first exploits the latent community structures to uncover the potential group-level patterns. Then, a dual-perspective community enhancement module is proposed, which strengthens the structural awareness and reinforces topological consistency within communities, thereby enabling more distinctive node representations and deeper intra-community message passing. Finally, a community-aware contrastive learning module is proposed, which considers nodes within the same community as positive pairs and those from different communities as negative pairs, enhancing the discriminability of node representations. Extensive experiments conducted on multiple benchmark datasets demonstrate that CAGCL consistently outperforms state-of-the-art baselines. The code is available at https://github.com/cgao-comp/.

With the expansion of business scales and scopes on online platforms, multi-scenario matching has become a mainstream solution to reduce maintenance costs and alleviate data sparsity. The key to effective multi-scenario recommendation lies in capturing both user preferences shared across all scenarios and scenario-aware preferences specific to each scenario. However, existing methods often overlook user-specific modeling, limiting the generation of personalized user representations. To address this, we propose PERSCEN, an innovative approach that incorporates user-specific modeling into multi-scenario matching. PERSCEN constructs a user-specific feature graph based on user characteristics and employs a lightweight graph neural network to capture higher-order interaction patterns, enabling personalized extraction of preferences shared across scenarios. Additionally, we leverage vector quantization techniques to distill scenario-aware preferences from users' behavior sequence within individual scenarios, facilitating user-specific and scenario-aware preference modeling. To enhance efficient and flexible information transfer, we introduce a progressive scenario-aware gated linear unit that allows fine-grained, low-latency fusion. Extensive experiments demonstrate that PERSCEN outperforms existing methods. Further efficiency analysis confirms that PERSCEN effectively balances performance with computational cost, ensuring its practicality for real-world industrial systems.

Multi-relational link prediction aims to forecast relationships among nodes within multi-relational graphs, with applications ranging from predicting drug interactions to completing knowledge graphs. Recent advancements have demonstrated that graph neural networks (GNNs), when augmented with heuristic information, significantly enhance the performance in this domain. However, existing approaches are limited by their uniform application of a single heuristic level across various datasets and often overlook the synergy between heuristic information and GNN architecture. Inspired by the successes of neural architecture search (NAS), this paper proposes a novel strategy that seeks to integrate multi-level heuristics with GNNs for enhanced multi-relational link prediction. This strategy involves a new framework that incorporates heuristic information at both global and local levels into GNNs, facilitating a cohesive application of these heuristics within the architecture. Building upon this framework, we have developed an extensive search space that includes widely-used heuristics and operations prevalent in manually designed architectures. Moreover, we employ a versatile search algorithm tailored to tackle the bi-level optimization challenge, ensuring efficient exploration of the search space. Empirical evaluations conducted on four benchmark datasets demonstrate that the proposed method significantly surpasses existing baselines in the task of multi-relational link prediction.

Subgraph-based methods have proven to be effective and interpretable in predicting drug-drug interactions (DDIs), which are essential for medical practice and drug development. Subgraph selection and encoding are critical stages in these methods, yet customizing these components remains underexplored due to the high cost of manual adjustments. In this study, inspired by the success of neural architecture search (NAS), we propose a method to search for data-specific components within subgraph-based frameworks. Specifically, we introduce extensive subgraph selection and encoding spaces that account for the diverse contexts of drug interactions in DDI prediction. To address the challenge of large search spaces and high sampling costs, we design a relaxation mechanism that uses an approximation strategy to efficiently explore optimal subgraph configurations. This approach allows for robust exploration of the search space. Extensive experiments demonstrate the effectiveness and superiority of the proposed method, with the discovered subgraphs and encoding functions highlighting the model’s adaptability.