A vast amount of user-generated content is uploaded on social media everyday, potentially being collected as training data for large language models, and thus poses severe threats to individual privacy. Existing approaches for protecting user-generated content often fail to detect implicit privacy leaks which are not directly given but could be inferred from the contextual text (clues). The precise detection of leaks and clues is difficult but critical in text sanitization. We propose a graph-based model for the user-generated content and build a privacy leakage detector PrivSniffer which integrates the inference capability of language models with the precise graph search, to capture the contextual privacy leakage (CPL). To evaluate the effectiveness of the framework, we create SynthLeak, a dataset of dialogues containing human-labeled implicit leaks, while featuring diversity and naturalness. Our experimental results on SynthLeak and several other benchmarks reveal PrivSniffer's superior capability in detecting implicit leaks and CPLs. Particularly in CPL detection, the state-of-the-art approach achieves only an F1 Score of 0.36, whereas PrivSniffer attains 0.59 on SynthLeak, showing great promise in real-world text sanitization.

Network alignment, pairing corresponding nodes across the source and target networks, plays an important role in many data mining tasks. Extensive studies focus on learning node embeddings across different networks in a unified space. However, these methods have not taken the large structural discrepancy between aligned nodes into account and, thus, are largely confined by the deterministic representations of nodes. In this work, we propose a novel network alignment framework highlighted by distributional learning and globally optimal alignment. By modeling the uncertainty of each node by Gaussian distribution, our framework builds similarity matrices on the Wasserstein distance between distributions and applies Sinkhorn operation, which learns the globally optimal mapping in an end-to-end fashion. We show that each integrated part of the framework contributes to the overall performance. Under a variety of experimental settings, our alignment framework shows superior accuracy and efficiency to the state-of-the-art.

The great power of Graph Neural Networks (GNNs) relies on a large number of labeled training data, but obtaining the labels can be costly in many cases. Graph Active Learning (GAL) is proposed to reduce such annotation costs, but the existing methods mainly focus on improving labeling efficiency with fixed classes, and are limited to handle the emergence of novel classes. We term the problem as Open-World Graph Active Learning (OWGAL) and propose a framework of the same name. The key is to recognize novel-class as well as informative nodes in a unified framework. Instead of a fully connected neural network classifier, OWGAL employs prototype learning and label propagation to assign high uncertainty scores to the targeted nodes in the representation and topology space, respectively. Weighted sampling further suppresses the impact of unimportant classes by weighing both the node and class importance. Experimental results on four large-scale datasets demonstrate that our framework achieves a substantial improvement of 5.97% to 16.57% on Macro-F1 over state-of-the-art methods.

Due to the universality of graph data, node classification shows its great importance in a wide range of real-world applications. Despite the successes of Graph Neural Networks (GNNs), GNN based methods rely heavily on rich connections and perform poorly on low-degree nodes. Since many real-world graphs follow a long-tailed distribution in node degrees, they suffer from a substantial performance bottleneck as a significant fraction of nodes is of low degree. In this paper, we point out that under-represented self-representations and low neighborhood homophily ratio of low-degree nodes are two main culprits. Based on that, we propose a novel method Grace which improves the node representation by self-distillation, and increases neighborhood homophily ratio of low-degree nodes by graph completion. To avoid error propagation of graph completion, label propagation is further leveraged. Experimental evidence has shown that our method well supports real-world graphs, and is superior in balancing degree-related bias and overall performance on node classification tasks.

We focus on the privacy-preserving problem in split learning in this work. In vanilla split learning, a neural network is split to different devices to be trained, risking leaking the private training data in the process. We novelly propose a patch shuffling scheme on transformers to preserve training data privacy, yet without degrading overall model performance. Formal privacy guarantees are provided and we further introduce the batch shuffling and the spectral shuffling schemes to enhance the guarantee. We show through experiments that our methods successfully defend the black-box, white-box, and adaptive attacks in split learning, with superior performance over baselines, and are efficient to deploy with negligible overhead compared to the vanilla split learning.

Recent studies have shown that graph neural networks (GNNs) are vulnerable to unnoticeable adversarial perturbations, which largely confines their deployment in many safety-critical domains. Robust graph structure learning has been proposed to improve the GNN performance in the face of adversarial attacks. In particular, the low-rank methods are utilized to purify the perturbed graphs. However, these methods are mostly computationally expensive with O(n3) time complexity and O(n2) space complexity. We propose LRGNN, a fast and robust graph structure learning framework, which exploits the low-rank property as prior knowledge to speed up optimization. To eliminate adversarial perturbation, LRGNN decouples the adjacency matrix into a low-rank component and a sparse one, and learns by minimizing the rank of the first part while suppressing the second part. Its sparse variant is formed to reduce the memory footprint further. Experimental results on various attack settings have shown LRGNN acquires comparable robustness with the state-of-the-art much more efficiently, boasting a significant advantage on large-scale graphs.