Steganographic threats are widespread in social networks, particularly on social media platforms. Existing methods mainly rely on statistical features and social network graph-based modeling, but struggle under sparsity and fragmentation due to limited semantics and weak graph connections that fail to capture deep inter-text dependencies. Therefore, we propose the CASS-LS framework, a Context-Aware and Semantic-Synergistic Linguistic Steganalysis framework for social network. CASS-LS consists of three key modules. The Semantic-Aware Feature Pretraining (SAFP) module generates more discriminative semantic representations to enhance the expressiveness of fragmented texts. The Semantic Synergy and Context-Aware Modeling (SSCA) module strengthens graph-based modeling to reconstruct inter-text relationships and capture long-range dependencies among sparse texts. The Semantics-Context-Synergistic Contrastive Loss Supervision (SCS-CLS) module compares semantic and contextual similarities between text pairs, improving model performance in detecting weak steganographic signals. Extensive experiments demonstrate that CASS-LS achieves superior F1 performance in social network steganalysis tasks.

Linguistic steganalysis aims to distinguish stego text from cover text. However, most existing methods heavily rely on a large number of stego text samples for training. In real-world scenarios, the cover text is far more abundant than the stego text, making it extremely difficult to obtain sufficient stego text for training. Furthermore, the scarcity of stego text also increases the difficulty of detection, posing greater challenges for steganalysis. In contrast, cover text is relatively easier to obtain in real-world scenarios, but current methods fail to fully utilize this resource. In this paper, we propose a Domain-Assisted Few-shot linguistic steganalysis method called DAF-Stega. To make full use of the cover text, we incorporate cover texts from multiple domains to assist in training. To address the scarcity of stego texts, we perform few-shot steganalysis based on a small amount of stego text and employ dynamic decision-making to generate pseudo-labels for self-training, enhancing model performance. Experimental results show that in few-shot learning scenarios, DAF-Stega effectively addresses the steganalysis problem under uncertain stego text proportions and outperforms existing methods.

Whether an outfit is compatible? Using machine learning methods to assess an outfit's compatibility, namely, fashion compatibility modeling (FCM), has recently become a popular yet challenging topic. However, current FCM studies still perform far from satisfactory, because they only consider the collocation compatibility modeling, while neglecting the natural human habits that people generally evaluate outfit compatibility from both the collocation (discrete assess) and the try-on (unified assess) perspectives. In light of the above analysis, we propose a Collocation and Try-On Network (CTO-Net) for FCM, combining both the collocation and try-on compatibilities. In particular, for the collocation perspective, we devise a disentangled graph learning scheme, where the collocation compatibility is disentangled into multiple fine-grained compatibilities between items; regarding the try-on perspective, we propose an integrated distillation learning scheme to unify all item information in the whole outfit to evaluate the compatibility based on the latent try-on representation. To further enhance the collocation and try-on compatibilities, we exploit the mutual learning strategy to obtain a more comprehensive judgment. Extensive experiments on the real-world dataset demonstrate that our CTO-Net significantly outperforms the state-of-the-art methods. In particular, compared with the competitive counterparts, our proposed CTO-Net significantly improves AUC accuracy from 83.2% to 87.8% and MRR from 15.4% to 21.8%. We have released our source codes and trained models to benefit other researchers.1