Multimodal-based recommendation integrates visual, textual, and acoustic modalities of items to comprehensively capture user preferences, thereby playing a crucial role in modern multimedia online platforms. In this paper, we propose aDiffusion-based Multi-agent withReinforcement Learning forMultimodal-basedRecommendation (DRMRec). Specifically, our DRMRec first leverages a hierarchicalDiffusion-basedMulti-Agent (DMA) to reconstruct the user-item interaction graph. Subsequently, these reconstructed graphs are fused with multimodal information to form contrastive views. During the reconstruction, each agent is injected with multimodal information through aCross-ModalAligner (CMA), thereby bringing cross-modal information closer to interaction embeddings and facilitating alignment across modalities. Then, we introduce aMetric-AwareDiffusionReinforcer (MADR), a reinforcement learning framework that leverages validation-set recommendation metrics as reward signals to enable dynamic and individual fine-tuning of each agent, thereby actively aligning model optimization with recommendation tasks. Next, we apply cross-modal graph contrastive learning to contrastive views, alleviating data sparsity while further enhancing cross-modal alignment. Extensive experiments on three real-world multimedia platform datasets demonstrate that DRMRec consistently outperforms state-of-the-art approaches in multimodal-based recommendation. It is noteworthy that the performance improvements across all three metrics on both the TikTok and Sports datasets exceed 10%.

Recommendation systems have been widely applied across various domains. However, they have consistently been hindered by the cold-start problem. One of the recent solutions involves extracting knowledge from other domains and applying it to the target domain’s recommendations. This is also one of the issues that cross-domain recommendation (CDR) aims to address. Mapping methods serve as a means to translate data across distinct domains, thereby boosting the performance of CDR. Meanwhile, diffusion probabilistic models can also be regarded as a superior data transformation model. They have achieved significant breakthroughs in image synthesis tasks, capable of recovering images from noise-added samples. To further improve the recommendation performance of CDR, we propose a novel CDR diffusion model [guided diffusion with bilateral alignment cross-domain recommendation (GDBACDR)]. First, a diffusion model (DM) is designed to guide the feature vectors of target domain users in the source and target domains. Simultaneously, a bilateral alignment module between the source and target domains is utilized to enhance the stability introduced by the DM, mitigate the negative impacts of randomness, and ensure the consistency of the reconstructed user embeddings with the target domain. The loss function of GDBACDR is formulated to achieve specific tasks. Finally, extensive experiments were conducted on existing datasets, and the results showed that the proposed method outperformed baseline models in every CDR task under the cold-start scenarios, demonstrating the effectiveness and adaptability of the proposed method.

Deep Neuro-Fuzzy Inference Systems (DNFIS) seamlessly fuse neural networks with the fuzzy inference system enabling intricate decision-making and knowledge representation, while upholding a commendable degree of adaptability and interpretability. However, the challenge of privacy-preserving inference (PI) over DNFIS has remained largely uncharted, with no prior research addressing this critical issue. In this paper, we embark on an exploration of this issue. We introduce an efficient and secure PI framework for DNFIS, named PrivDNFIS, which leverages the post-quantum lattice-based homomorphic encryption to implement secure computation protocols for PI over DNFIS. Our work incorporates several non-trivial performance enhancements. Firstly, it consolidates multiple elements of input feature vectors into a single message, reducing encryption/decryption overhead. Secondly, building upon this novel encoding approach, PrivDNFIS can perform ciphertext aggregation and vector-vector inner production without necessitating time-consuming ciphertext rotation operations. Thirdly, we replace the softmax function in the DNFIS layer with a quadratic function to further enhance inference efficiency, without compromising the inference accuracy. Under the given threat model, we provide formal security proof for PrivDNFIS. In comprehensive experimental results, PrivDNFIS demonstrates an approximately 1.9 to 4.4 times reduction in end-to-end time cost compared to the benchmark.

Social network alignment has made significant progress in social network analysis, with representative applications such as cross-domain recommendation and community detection. However, existing approaches require institutions to share raw user data, raising significant privacy concerns. To address this issue, we propose a Privacy-Preserving Network Alignment (PPNA) scheme that eliminates the need for raw data sharing. In concrete, PPNA leverages homomorphic encryption to enable computation over the ciphertext domain without decryption. It ensures provable data privacy. PPNA also presents a secure multiparty computation protocol to eliminate reliance on trusted third-party servers, which is often impractical in real-world scenarios. Furthermore, its well-designed iterative update mechanism is well-suited for iterative alignment algorithms. Comprehensive experimental results have demonstrated that PPNA improves performance compared to the scenario where raw data sharing is unfeasible due to privacy concerns. It achieves an average F1-score increase of 1.65 times and up to 2.28 times. The performance gain is more pronounced in decentralized settings, highlighting PPNA’s practicality in real-world scenarios when multi-institution collaboration is imperative.

Large language models (LLMs) can transform natural language instructions into executable commands for IoT devices like unmanned aerial vehicles (UAVs), creating new development opportunities. However, safety concerns about LLMs translating commands into machine or program control instructions cannot be overlooked. Currently, jailbreak instructions used to test the LLM security are often restricted to specific modes or tasks, resulting in a lack of diversity and leaving some tasks unexplored. To address this issue, we introduce a Multi-Instruction Fusion (MIF) method that can automatically fuse harmful prompts and various task instructions into jailbreaks. Firstly, we adopt a reverse decomposition strategy to acquire sufficient supervised data for fusing harmful prompts and instructions into jailbreaks and construct a task instruction synthesizer based on it. Then, to determine the optimal instruction combinations in the vast combination space, we propose a representative-node-based selection strategy, ReNB, to rank and filter the instruction combinations on a few representative samples, thereby accelerating the identification of the valid ones. Experimental results demonstrate that MIF significantly improves the attack success rate, achieving over 90% on GPT-4o-mini, LLaMa2-70B and Qwen2-7B models, outperforming the state-of-the-art baselines.

Multi-relational graph learning aims to embed entities and relations in knowledge graphs into low-dimensional representations, which has been successfully applied to various multi-relationship prediction tasks, such as information retrieval, question answering, and etc. Recently, contrastive learning has shown remarkable performance in multi-relational graph learning by data augmentation mechanisms to deal with highly sparse data. In this paper, we present a Multi-Relational Graph Contrastive Learning architecture (MRGCL) for multi-relational graph learning. More specifically, our MRGCL first proposes a Multi-relational Graph Hierarchical Attention Networks (MGHAN) to identify the importance between entities, which can learn the importance at different levels between entities for extracting the local graph dependency. Then, two graph augmented views with adaptive topology are automatically learned by the variant MGHAN, which can automatically adapt for different multi-relational graph datasets from diverse domains. Moreover, a subgraph contrastive loss is designed, which generates positives per anchor by calculating strongly connected subgraph embeddings of the anchor as the supervised signals. Comprehensive experiments on multi-relational datasets from three application domains indicate the superiority of our MRGCL over various state-of-the-art methods. Our datasets and source code are published at https://github.com/Legendary-L/MRGCL.

The importance of social network alignment (SNA) for various downstream applications, such as social network information fusion and e-commerce recommendation, has prompted numerous professionals to develop and share SNA tools. However, malicious actors can exploit these tools to integrate sensitive user information, thereby posing cybersecurity risks. Although many researchers have explored attacking SNA (ASNA) through network modification attacks to protect users, practical feasibility remains challenging. In this study, we propose an effective node injection attack via a dynamic programming framework (DPNIA) to address the problem of modeling and solving ASNA within a limited time and balancing the costs and benefits. DPNIA models ASNA as a problem of maximizing the number of confirmed incorrect correspondent node pairs with greater similarity scores than the pairs between existing nodes, thereby making ASNA solvable. A cross-network evaluation method is employed directly to identify node vulnerabilities, facilitating progressive attacking from easy to difficult. In addition, an optimal injection strategy searching method based on dynamic programming is used to determine which links should be added between the injected and existing nodes, thereby enhancing the effectiveness of the attack at a low cost. Experiments on four real-world datasets demonstrated that DPNIA consistently and significantly surpasses various baselines when attacking both multiple networks simultaneously and a single network.

Multimedia recommendations aim to use rich multimedia content to enhance historical user-item interaction information, which can not only indicate the content relatedness among items but also reveal finer-grained preferences of users. In this paper, we propose aKnowledge-awareDiffusion-Enhanced architecture using contrastive learning paradigms (KDiffE) for multimedia recommendations. Specifically, we first utilize original user-item graphs to build an attention-aware matrix into graph neural networks, which can learn the importance between users and items for main view construction. The attention-aware matrix is constructed by adopting a random walk with a restart strategy, which can preserve the importance between users and items to generate aggregation of attention-aware node features. Then, we propose a guided diffusion model to generate strongly task-relevant knowledge graphs with less noise for constructing a knowledge-aware contrastive view, which utilizes user embeddings with an edge connected to an item to guide the generation of strongly task-relevant knowledge graphs for enhancing the item's semantic information. We perform comprehensive experiments on three multimedia datasets that reveal the effectiveness of our KDiffE and its components on various state-of-the-art methods. Our source codes are available.

The pervasiveness of offensive language on social networks has caused adverse effects on society, such as abusive behavior online. It is urgent to detect offensive language and curb its spread. In the popular datasets, the distribution of users and tweets is imbalanced, which limits the generalization ability of the model. In addition, existing research shows that methods with community information extracted from the social graphs effectively improve the performance of offensive language detection. However, the existing models deal with social graphs independently, which seriously affects the effectiveness of detection models. In this article, we release a new dataset with users and social relationships. To encode community information, we construct the social graphs based on the user historical behavior information and social relationships. Moreover, we propose a model based on graph attention networks (GATs) and fusion features for offensive language detection (GF-OLD). Specifically, the community information is directly captured by the GAT module, and the text embeddings are taken from the last hidden layer of bidirectional encoder representation from transformer (BERT). Attention mechanisms and position encoding are used to fuse these features. Our method outperforms baselines with the F1-score of 89.94%. The results show that our model effectively learns the potential information of social graphs and text, and user historical behavior information is more suitable for user attribute in the social graphs.

Online users are typically active on multiple social media networks (SMNs), which constitute a multiplex social network. With improvements in cybersecurity awareness, users increasingly choose different usernames and provide different profiles on different SMNs. Thus, it is becoming increasingly challenging to determine whether given accounts on different SMNs belong to the same user; this can be expressed as an interlayer link prediction problem in a multiplex network. To address the challenge of predicting interlayer links, feature or structure information is leveraged. Existing methods that use network embedding techniques to address this problem focus on learning a mapping function to unify all nodes into a common latent representation space for prediction; positional relationships between unmatched nodes and their common matched neighbors (CMNs) are not utilized. Furthermore, the layers are often modeled as unweighted graphs, ignoring the strengths of the relationships between nodes. To address these limitations, we propose a framework based on multiple types of consistency between embedding vectors (MulCEVs). In MulCEV, the traditional embedding-based method is applied to obtain the degree of consistency between the vectors representing the unmatched nodes, and a proposed distance consistency index based on the positions of nodes in each latent space provides additional clues for prediction. By associating these two types of consistency, the effective information in the latent spaces is fully utilized. In addition, MulCEV models the layers as weighted graphs to obtain representation. In this way, the higher the strength of the relationship between nodes, the more similar their embedding vectors in the latent representation space will be. The results of our experiments on several real-world and synthetic datasets demonstrate that the proposed MulCEV framework markedly outperforms current embedding-based methods, especially when the number of training iterations is small.

User identification can help us build more comprehensive user information. It has been attracting much attention from academia. Most of the existing works are profile-based user identification and relationship-based user identification. Due to user privacy settings and social network restrictions on user data crawl, user data may be missing or incomplete in real social networks. User data include profiles, user-generated contents (UGCs), and relationships. The features extracted in previous research may be sparse. In order to reduce the impact of the above problems on user identification, we propose a multiple user information user identification framework (MUIUI). Firstly, we develop multiprocess crawlers to obtain the user data from two popular social networks, Twitter and Facebook. Secondly, we use named entity recognition and entity linking to obtain and integrate locations and organizations from profiles and UGCs. We also extract URLs from profiles and UGCs. We apply the locations jointly with the relationships and develop several algorithms to measure the similarity of the display name, all locations, all organizations, location in profile, all URLs, following organizations, and user ID, respectively. Afterward, we propose a fusion classifier machine learning-based user identification method. The results show that the F1 score of MUIUI reaches 86.46% on the dataset. It proves that MUIUI can reduce the impact of user data that are missing or incomplete.