Modern and effective recommender systems leverage not only user-item interactions but also private attributes of users to bring promising performance. Protecting private attributes from being inferred by the adversary has become a vital issue in recommender systems. In this work, we formulate the problem of privacy-preserving representation learning for recommendation (PrP-Rec). The design of PrP-Rec is to generate embeddings of users and items so that two inference attacks can be effectively defended. One is item-based attribute inference attack (IAI-Attack), and the other is embedding-based attribute retrieval attack (EAR-Attack). To tackle the PrP-Rec problem, we present a novel framework, privacy-preserving Bayesian personalized ranking (PBPR). The key is to create a learnable gradient obfuscation vector and have it injected into the embedding learning of users and items. The objective of gradient obfuscation is devised to optimize with recommendation and privacy protection. Extensive experiments conducted on three benchmark datasets exhibit that PBPR can outperform competing methods of privacy-preserving recommendation in the top-K recommendation and effectively defending IAI-Attack and EAR-Attack.

Graph Neural Networks (GNNs) have recently achieved remarkable performance on the node classification task. While most typical GNN models presume that the graph data is clean, however, graphs could be polluted by various noises that hurt the prediction accuracy. Besides, while GNNs rely on sufficient labeled data to propagate the supervision signal, we should move to a more realistic setting—to learn with a limited amount of labeled data, i.e., label scarcity. In this work, we aim at building a Holistically Robust Graph Neural Network (HRGNN) against four different types of graph noise, including adversarial attacks, edge sparsity, noisy labels, and high heterophily, in the presence of label scarcity. We proposed a novel GNN framework, HRGNN, to fulfill the goal. The main idea of HRGNN is to create synthetic nodes with labels and learn to properly connect them with existing nodes. With synthetic nodes, HRGNN can inject reliable information into existing nodes due to the message-passing mechanism of GNNs that helps purify the polluted representations of nodes and alleviate the negative effect caused by various noises. Furthermore, edge filtering in HRGNN helps remove the noisy edges to prevent the propagation of incorrect information, while pseudo-labeling provides more label information to defend against label scarcity and label noise. Experiments conducted on eight graph datasets exhibit that HRGNN consistently outperforms the state-of-the-art GNN competing models on four types of noisy settings with label scarcity. To the best of our knowledge, HRGNN is the first GNN model that is holistically robust to various types of noise and label scarcity.

Recent academic research reveals that strategic information operations often co-opt authentic communities to produce participatory propaganda—the manufacturing, curation, and amplification of stories that advance partisan narratives. In this paper, we push the field to move beyond disinformation and rhetorical techniques to a more productive narrative-based approach to propaganda analysis, and propose the first systematic computational method for the study of participatory propaganda. We propose a theory-driven computational framework centered on the concept of the propaganda trope. Tropes are recurring scenes or stories that effectively anchor and communicate persuasive narratives. We argue that tropes provide a blueprint for impactful propaganda, and offer a valuable analytical lens for understanding how participatory propaganda is created and sustained. We illustrate this argument through an empirical case study examining state media and a state-aligned partisan community on Twitter.

Handling noisy labels during neural network training is fundamental, as such noise can skew model optimization, leading to diminished performance. This article zeroes in on node classification using graph neural networks (GNNs), renowned for their adeptness in harnessing both node attributes and edge connections. Nevertheless, GNNs, due to their inherent feature propagation and information amalgamation, are more susceptible to misdirection from label noise than conventional neural networks. While existing robust GNNs offer defenses against noise’s adverse effects, they often underrate risks posed by noisy labels during training or undervalue the significance of fortifying edges against erroneous aggregation and noisy edges. Addressing these gaps, we introduce a novel robust GNN framework, noise-mitigating GNN (NomiGNN), designed to enhance the robustness of GNNs against label noise. NomiGNN estimates noise distributions and refines loss optimization to counteract noise during training. Moreover, by instituting edge labels for a fresh prediction task, we facilitate learning sample relationships via same-label probabilities, mitigating mis-aggregation from noisy edges and bolstering node classification. Supplementing this, we incorporate pseudoedge labeling and iterative learning to remedy label shortages and estimation inaccuracies. NomiGNN is substantiated by both theoretical rationale and empirical evidence. Experimental evaluations on five real-world graphs highlight NomiGNN’s superior resilience against noisy label corruption, outperforming eight benchmark GNN models.

Research on state-sponsored strategic information operations has emphasized detection, while reporting and attribution—particularly the tagging of tactics, techniques, and procedures (TTPs)—remain underexplored. Accurate TTP tagging of incidents facilitates attribution through forensic evidence, but progress is constrained by scarce datasets, inconsistent ontologies, and a long tail of low-frequency TTPs. Existing resources are limited in annotator diversity, domain scope, and incident-level disaggregation, restricting their utility. In this work, we present the first incident-level dataset of public reports annotated with complete sets of TTPs, with gold labels established through consensus among multiple expert annotators. Our dataset emphasizes cases from the Indo-Pacific region, addressing a significant geographic gap in existing resources, and is accompanied by our codebook that augments the description of the ontology. We evaluate generative large language models (LLMs) on this dataset, finding peak performance only at micro-F1 of.669, but good enough as a semi-automated system to help with high levels of false negatives in human annotation. We investigate key failure modes that impede reliable automation. Our analysis identifies how annotation guidelines, cost-performance trade-offs, and systematic error patterns shape LLM performance, offering immediate directions for future model design and human-in-the-loop approaches to TTP tagging.

Real-world analytics hinges on tabular data, yet prevailing learners face a triple bind: tree ensembles excel on fixed schemas but cannot generalize to new columns, neural nets learn rich features yet overfit small tables, and recent transfer approaches falter when schemas diverge. We tackle these limitations with GTab, a Gradient-Boosting Bipartite Graph Neural Network that marries decision-tree residual refinement with self-supervised graph representation learning. GTab maps each table to an instance-feature bipartite graph, where a GNN, optimised jointly with contrastive, clustering, and reconstruction objectives, captures feature-feature, instance-instance, and cross-type relations. Boosted trees ingest the GNN's gradients, correcting residual errors and injecting the strong inductive bias of split-based models. Building on this backbone, we introduce three variants: E-GTab ensembles multiple overlapping feature sub-graphs for robust classic prediction; I-GTab inductively attaches unseen feature nodes, enabling feature-incremental inference without retraining; and T-GTab pre-trains on a source schema and lightly fine-tunes on a target schema to achieve zero-shot and transfer learning across heterogeneous tables. Across 20 public benchmarks and five clinical trials, GTab consistently ranks first: it outperforms tree, neural, and graph baselines on static tasks, surpasses prior art (and an oracle) when half the test-time columns are unseen, and delivers higher AUC than the leading transformer baseline in both cross-dataset and zero-shot transfer - all with a unified architecture. GTab thus offers a principled, scalable, and adaptable solution to holistic tabular data prediction, bridging the gap between classic ensembles and modern self-supervised representation learning.

Graph neural networks (GNNs) have become the de facto standard for learning over graph-structured data, yet they often falter when labels are scarce, graphs are noisy, or structural patterns are complex. While boosting is a powerful ensemble paradigm that sequentially corrects model errors, existing adaptations to graphs either break end-to-end training by introducing external boosters or resort to explicit sample reweighting that undermines relational inductive biases. To overcome these challenges, we introduce adaptive-boosting graph neural networks (AdaGNN), a fully integrated boosting framework designed specifically for graph learning. AdaGNN trains a series of GNN learners in a unified pipeline: each learner is warm-started from the previous one and ingests the progressively refined node embeddings. After each round, correctly classified nodes are frozen, i.e., their embeddings locked and excluded from subsequent updates, so that remaining learners focus solely on residual errors. This Embedding Freeze mechanism implicitly propagates error signals through the graph topology without requiring explicit weight adjustments or multi-stage coordination. Comprehensive evaluations across varied supervision levels demonstrate that AdaGNN consistently elevates classification accuracy and robustness, delivering substantial gains under limited labels while preserving or enhancing performance when labels are plentiful. AdaGNN establishes a novel, graph-native adaptive boosting paradigm—melding sequential GNN refinement, implicit error handling, and end-to-end training—to significantly advance the accuracy and robustness of GNN applications.

In this survey, we dive into Tabular Data Learning (TDL) using Graph Neural Networks (GNNs), a domain where deep learning-based approaches have increasingly shown superior performance in both classification and regression tasks compared to traditional methods. The survey highlights a critical gap in deep neural TDL methods: the underrepresentation of latent correlations among data instances and feature values. GNNs, with their innate capability to model intricate relationships and interactions between diverse elements of tabular data, have garnered significant interest and application across various TDL domains. Our survey provides a systematic review of the methods involved in designing and implementing GNNs for TDL (GNN4TDL). It encompasses a detailed investigation into the foundational aspects and an overview of GNN-based TDL methods, offering insights into their evolving landscape. We present a comprehensive taxonomy focused on constructing graph structures and representation learning within GNN-based TDL methods. In addition, the survey examines various training plans, emphasizing the integration of auxiliary tasks to enhance the effectiveness of instance representations. A critical part of our discussion is dedicated to the practical application of GNNs across a spectrum of GNN4TDL scenarios, demonstrating their versatility and impact. Lastly, we discuss the limitations and propose future research directions, aiming to spur advancements in GNN4TDL. This survey serves as a resource for researchers and practitioners, offering a thorough understanding of GNNs' role in revolutionizing TDL and pointing towards future innovations in this promising area.

This article introduces the Missingness-aware Self-Supervised Graph Neural Network (MissGNN), a novel model for tabular data predictions in scenarios with missing values. Addressing a significant challenge in fields like finance, healthcare, and environmental studies, MissGNN innovatively integrates feature imputation and label prediction within a unified framework. Utilizing a Tabular Bipartite Graph, it distinctively represents data points and features, enabling intricate modeling of their relationships. This approach is pivotal in contexts where missing data can significantly skew model training and predictions. MissGNN employs a Graph Neural Network to generate embeddings, capturing interactions between data points and features for accurate missing value imputation and label prediction. It features a dual-focus mechanism, comprising edge-wise feature imputation and node-wise label prediction. A pre-training phase for feature imputation enhances the model’s predictive ability for both features and labels. Additionally, MissGNN incorporates a multi-view learning strategy through feature subsetting, further enriching its representation learning. Experiments on 14 tabular datasets demonstrate MissGNN’s superiority over existing models in managing missing values across both regression and classification tasks. The results underscore its robustness and practical applicability, establishing it as a versatile tool for tackling tabular data prediction challenges with missing values. MissGNN’s contributions significantly redefine the approach to missing value imputation, aligning feature imputation and label prediction in a cohesive manner and underscoring the potential of Graph Neural Networks in processing tabular data.

Recommender systems require on-demand unlearning of user-item interactions to meet privacy regulations without sacrificing recommendation quality. Existing approaches either retrain large models at prohibitive cost or apply local parameter tweaks that misestimate non-linear effects, resulting in degraded accuracy, poor scalability for bulk deletions, and no formal privacy auditing. We present ReCUR, a Recommendation Contrastive Unlearning framework with Influence Estimation that addresses these weaknesses. ReCUR first applies a contrastive push-pull loss to push forgotten interaction embeddings away and pull retained ones back toward their original anchors. This not only subtracts deleted interactions’ contributions from the model, but yields fine-grained forgetting control, minimizes collateral drift, and recovers lost utility without full retraining. To compensates for any lost utility and promote fairness, ReCUR introduces a re-ranking mechanism with group-specific promotion weights that rebalance recommendations for diverse and niche user segments. Across four real-world datasets, ReCUR matches or exceeds retrain-from-scratch accuracy, retaining over 95

Online clickbait continues to plague social-media platforms, where sensational captions lure users into low-value or misleading content. While prior work has explored individual modeling choices, i.e., sequential encoders, graph-based representations, and simple fusion strategies, no study has systematically compared these design dimensions in the clickbait domain. We address this gap by conducting the first comprehensive analysis of three core axes: (1) how to organize the model streams (treating caption and hashtags jointly vs. separately), (2) how to learn text representations (sequential vs. graph-based), and (3) how to fuse these modalities (concatenation vs. co-attention). Leveraging a large, manually labeled Instagram dataset of short captions paired with hashtags, we implement every combination of these axes to isolate their individual and joint impacts on detection performance. Our experiments reveal clear trends: processing caption and hashtags in parallel streams preserves their distinct semantic patterns and consistently outperforms unified processing; graph-based embeddings capture long-range and corpus-wide co-occurrence structures that sequential models alone miss; and a co-attention fusion mechanism aligns caption and hashtag signals, uncovering subtle mismatches characteristic of clickbait.

This work presents a novel approach to tabular data prediction leveraging graph structure learning and graph neural networks. Despite the prevalence of tabular data in real-world applications, traditional deep learning methods often overlook the potentially valuable associations between data instances. Such associations can offer beneficial insights for classification tasks, as instances may exhibit similar patterns of correlations among features and target labels. This information can be exploited by graph neural networks, necessitating robust graph structures. However, existing studies primarily focus on improving graph structure from noisy data, largely neglecting the possibility of deriving graph structures from tabular data. We present a novel solution, Tabular Graph Structure Learning (TabGSL), to enhance tabular data prediction by simultaneously learning instance correlation and feature interaction within a unified framework. This is achieved through a proposed graph contrastive learning module, along with transformer-based feature extractor and graph neural network. Comprehensive experiments conducted on 30 benchmark tabular datasets demonstrate that TabGSL markedly outperforms both tree-based models and recent deep learning-based tabular models. Visualizations of the learned instance embeddings further substantiate the effectiveness of TabGSL.

Mental health issues have become a growing global concern, especially with the increasing expression of emotions and distress on social media platforms. Effectively leverage NLP techniques to identify user with potential suicidal tendencies is crucial for providing timely prevention. In this paper, we propose the Prediction of Writing-based Emotional Risk with Mental-adaptive Multi-modal Learning (PoWER-M) model in tackling the challenge in the IEEE BigData 2025 Cup: Suicide Risk Prediction on Social Media. Given a sequence of five user-level posts on Reddit, the task is to predict the suicide risk level of subsequent post. We highlight the challenges and insights in this suicide-risk prediction task: (i) severe imbalance among the categories of risk, (ii) the crucial role of temporal features in emotional state shifts, and (iii) the importance of domain alignment within mental field. To address these issues, we introduce a two-stage architecture, integrating MentalRoBERTa for domain-specific contextual representation and a Bidirectional-LSTM for modeling inter-post dependencies and extracted temporal dynamics. Furthermore, a hybrid loss function is designed to alleviate the minority bias of the high-risk category Attempt. Lastly, the framework is further enhanced with ensemble strategy to improve robustness and performance. Extensive experiments show that PoWER-M outperforms strong baselines including state-of-the-art tree models and transformers. Ablation study highlight the contribution of temporal features, domain-aligned language modeling, and sequential attention, demonstrating that multi-modeling significantly enhances detection of high-risk posts. The PoWER-M model ultimately achieved second place on the private leaderboard, demonstrating the effectiveness of combining domain-specific language model with temporal dynamics for suicide risk prediction.

Real-world anomaly detection seldom enjoys abundant, perfectly curated normal data; labels are often limited and untrustworthy, allowing genuine outliers to masquerade as benign. We cast oneclass anomaly detection into this harsh setting of label scarcity and noise and present Random-Feature Graph Neural Networks with a Representation Tokenized Transformer (RFGT). RFGT first partitions the feature space into random, non-overlapping subsets, yielding multiple complementary views that dilute the impact of any corrupted dimension. With a graph constructed from each view, a graph learner propagates the sparse normal signal to unlabeled neighbors while tempering mislabeled anomalies. A novel Representation Tokenized Transformer module is learned to capture cross-feature dependencies and implicitly down-weight inconsistent signals within a data instance. Extensive experiments on eight tabular benchmarks exhibit that RFGT outperforms the state-of-the-art as the amount of clean data shrinks or the label noise grows.

The rapid evolution of Large Language Models (LLMs) has pro-foundly impacted social media, transforming how information is generated, disseminated, and analyzed. With their ability to process vast amounts of data, grasp contextual nuances, and engage in human-like dialogue, LLMs present new opportunities and challenges for understanding online interactions. The SocialLLM 2025 workshop builds on the foundation laid by SocialNLP, expanding the scope to explore the capabilities and implications of LLMs in social media research. This year's workshop at TheWebConf 2025 focuses on three pivotal themes: leveraging LLMs for mental health support, enhancing emotion detection in textual interactions, and improving misinformation detection through active learning. The selected papers illustrate cutting-edge advancements in these areas, demonstrating how LLMs can be fine-tuned for therapeutic dialogue generation, assessed for their emotional intelligence, and optimized for misinformation detection with minimal labeled data. These contributions highlight the growing interdisciplinary nature of LLM research, merging insights from natural language processing, social computing, and artificial intelligence ethics. By bringing together researchers and practitioners from diverse backgrounds, SocialLLM 2025 aims to foster meaningful discussions on the opportunities and risks associated with LLM-driven social media applications. The workshop serves as a platform for exploring novel methodologies, addressing ethical concerns, and shaping future directions for responsible AI deployment in social media environments. Through collaborative efforts, we seek to advance the field and ensure that LLMs contribute positively to the digital ecosystem.

Learning from data with missing values is a common challenge in real-world applications. Existing approaches for handling data incompleteness often involve imputation, which can introduce errors that propagate into downstream tasks or impose assumptions that limit the support for heterogeneous feature types. To address these issues, we propose Missing Feature Attention Network ( MFAN ), an end-to-end label prediction model that directly consumes incomplete data without requiring imputation. MFAN flexibly accommodates both continuous and categorical features through learnable embeddings, and leverages a transformer encoder with self-attention to capture the correlation among features as well as the correlation between features and missingness . This attention-based mechanism allows missing features to benefit from relationships learned among observed features, leading to enhanced hidden representations and robust prediction performance. Additionally, we introduce auxiliary self-supervised pre-training tasks that further guide the attention mechanism in modeling missingness. Experimental results on eight regression and seven classification datasets demonstrate MFAN ’s superiority over state-of-the-art end-to-end methods and imputation-based approaches. Comprehensive ablation studies confirm the effectiveness of each MFAN component, underscoring the importance of explicitly modeling correlations among observed and missing features.

Deceptive content, from social bots to fake hotel reviews, erodes trust in markets, media, and democratic discourse. Detecting it is an urgent AI-for-social-good problem, yet real deployments confront three hurdles that current models handle only in isolation: scarce and noisy labels, and extreme class imbalance. We present Generative Text Graph Neural Networks (GetGNN), the first framework to address all three simultaneously. GetGNN embeds words and documents in a heterogeneous graph, then generates new minority documents by interpolating deceptive embeddings and predicts their edges via a self-supervised link reconstructor. A joint objective couples edge generation with classification, letting the model propagate limited supervision while diluting corrupted labels. Extensive experiments on the PAN-19 Twitter-bot and Spam hotel-review benchmarks show that GetGNN significantly boosts detection performance over state-of-the-art baselines under every combination of label scarcity and noise. By providing a principled, data-efficient, and noise-resilient detector, GetGNN advances the practical safeguarding of online ecosystems for the public good.

Graph neural recommenders excel at propagating collaborative signals, yet they still depend on vanishingly sparse user-item interactions. Existing solutions inject auxiliary losses or hand-crafted perturbations, but the underlying supervision remains unchanged and the sparsity barrier persists. We propose PARIS, a Predict-Add-Refine Scheme that reconceives training as an iterative densification process. In each round, the model first predicts the unseen items it is most confident about, then permanently adds those links to the graph, and finally retrains on the enriched structure. This loop repeats until additional passes yield no measurable lift, turning sparsity into a self-regulated curriculum. We prove that every generation strictly reduces pairwise ranking loss under benign label noise and that the marginal gain grows as the original graph becomes sparser, providing the formal link between supervision density and graph-neural optimization. PARIS is architecture-agnostic, so any modern backbone can inherit its benefits without new parameters or hyperparameters. Extensive experiments across diverse domains confirm the theory, showing consistent performance improvement, and disproportionate gains for cold-start users. Our findings indicate that intelligent supervision densification, rather than sophisticated regularization, may be all one needs for substantial progress in graph-based recommendation.

Early detection of patient deterioration is essential for timely treatment, with vital signs like heart rates being key health indicators. Existing methods tend to solely analyze vital sign waveforms, ignoring transition relationships of waveforms within each vital sign and the correlation strengths among various vital signs. Such studies often overlook nuanced illness deterioration, which is the early sign of worsening health but is difficult to detect. In this paper, we introduce CAND, a novel method that organizes the transition relationships and the correlations within and among vital signs as domain-specific and cross-domain knowledge. CAND jointly models these knowledge in a unified representation space, considerably enhancing the early detection of nuanced illness deterioration. In addition, CAND integrates a Bayesian inference method that utilizes augmented knowledge from domain-specific and cross-domain knowledge to address the ambiguities in correlation strengths. With this architecture, the correlation strengths can be effectively inferred to guide joint modeling and enhance representations of vital signs. This allows a more holistic and accurate interpretation of patient health. Our experiments on a real-world ICU dataset demonstrate that CAND significantly outperforms existing methods in both effectiveness and earliness in detecting nuanced illness deterioration. Moreover, we conduct a case study for the interpretable detection process to showcase the practicality of CAND.

Sharing tabular data often requires removing or replacing sensitive attributes, yet existing approaches rarely meet three requirements at once: preserve downstream task performance, match real-world feature distributions, and limit privacy leakage. We present Bipartite Tabular Graph Neural Networks (BitGNN) for Privacy-preserving Data Synthesis, a unified framework that treats a table as a bipartite graph with observation nodes, feature nodes, and edge attributes holding cell values. BitGNN learns to synthesize a sensitive column as an edge-level prediction task while simultaneously optimizing label prediction as a node-level task. The model is pre-trained by randomly masking cell values and reconstructing them, then finetuned end-to-end for the downstream prediction objective. At release time, BitGNN removes the true sensitive column, samples its replacement conditionally on the non-sensitive features with independent draws per record, and applies a light marginal calibration, yielding privacy by design via correlation control and strong distribution fidelity. On two public regression datasets, BitGNN consistently outperforms widely used imputation baselines and approaches the performance of models trained on original data, while better matching the marginals of the sensitive attribute and preserving only the dependence mediated by nonsensitive features. The method delivers a practical path to shareable, task-useful, and privacy-respecting tabular data with a single coherent model and a reproducible evaluation protocol.