Integrating multimodal data of pathological image and gene expression for cancer survival analysis can achieve better results than using a single modality. However, existing multimodal learning methods ignore fine-grained interactions between both modalities, especially the interactions between biological pathways and pathological image patches. In this article, we propose a novel Pathway-Aware Multimodal Transformer (PAMT) framework for interpretable cancer survival analysis. Specifically, the PAMT learns fine-grained modality interaction through three stages: (1) In the intra-modal pathway-pathway / patch-patch interaction stage, we use the Transformer model to perform intra-modal information interaction; (2) In the inter-modal pathway-patch alignment stage, we introduce a novel label-free contrastive loss to aligns semantic information between different modalities so that the features of the two modalities are mapped to the same semantic space; and (3) In the inter-modal pathway-patch fusion stage, to model the medical prior knowledge of "genotype determines phenotype", we propose a pathway-to-patch cross fusion module to perform inter-modal information interaction under the guidance of pathway prior. In addition, the inter-modal cross fusion module of PAMT endows good interpretability, helping a pathologist to screen which pathway plays a key role, to locate where on whole slide image (WSI) are affected by the pathway, and to mine prognosis-relevant pathology image patterns. Experimental results based on three datasets of bladder urothelial carcinoma, lung squamous cell carcinoma, and lung adenocarcinoma demonstrate that the proposed framework significantly outperforms the state-of-the-art methods.

Conversational Recommender Systems (CRS) enable more accurate and personalized recommendation through multi-turn interactions by dynamically capturing user preferences. However, the data imbalance in CRSs, due to the inconsistent distribution of user interaction histories, may lead to biased treatment for disadvantaged user group. This paper systematically investigates the discriminate problems in CRS from the user's perspective, called as User-Oriented Fairness (UOF). Through comprehensive empirical analysis across multiple dimensions, we reveal significant disparities in recommendation quality and representation learning among user groups with varying interaction frequencies. To address these challenges, we propose a user fairness enhancement framework, named DareCRS, which is a model-agnostic framework. The novelty of DareCRS lies in two key components: (1) At the data level, it integrates LLM-powered Dialog Augmentation with a hybrid sampling strategy to effectively mitigate data sparsity for disadvantaged users; (2) At the representation level, it introduces a Representation Equalization mechanism based on optimal transport between user groups, aiming to enhance the embedding space of disadvantaged users via knowledge transfer. Extensive experiments on two public datasets show that DareCRS achieves an average improvement of 5.37% in recommendation performance and an average enhancement of 43.14% in recommendation fairness across user groups compared to current state-of-the-art baseline methods. Our code and data are available at: https://anonymous.4open.science/r/DareCRS-8B94.

Deep learning based recommender systems(DLRS) as one of the up-and-coming recommender systems, and their robustness is crucial for building trustworthy recommender systems. However, recent studies have demonstrated that DLRS are vulnerable to data poisoning attacks. Specifically, an unpopular item can be promoted to regular users by injecting well-crafted fake user profiles into the victim recommender systems. In this paper, we revisit the data poisoning attacks on DLRS and find that state-of-the-art attacks suffer from two issues: user-agnostic and fake-user-unitary or target-item-agnostic, reducing the effectiveness of promotion attacks. To gap these two limitations, we proposed our improved method Generate Targeted Attacks(GTA), to implement targeted attacks on vulnerable users defined by user intent and sensitivity. We initialize the fake users by adding seed items to address the cold start problems of fake users so that we can implement targeted attacks. Our extensive experiments on two real-world datasets demonstrate the effectiveness of GTA.

The federated framework is actively applied in knowledge graph fusion research to obtain a complete knowledge graph without exposing data privacy. It can help local clients learn the knowledge graph embeddings in other clients without revealing data privacy. However, current federated-based knowledge graph embedding frameworks cannot exploit both entity and relation embeddings and may not prevent partial triples from being reconstructed. This paper proposes a novel framework named Federated Multi-server knowledge graph embedding (FedM), which creatively utilizes uploaded entity and relation embeddings while preventing privacy leakage. Expressly, we first set up two central servers for entity and relation embeddings to aggregate and share client-uploaded embeddings. Secondly, we design a knowledge graph secure aggregation algorithm to address the potential privacy concerns in FedM. We conduct comparative experiments on an empirical dataset (divided into three federated datasets) with four commonly-used knowledge graph embedding methods to evaluate the performance of our proposed framework. In addition, our proposed FedM framework is generally superior to the latest baseline frameworks on both privacy preservation and link prediction tasks.

Point of Interest (POI) recommendation system rec-ommends places in which users have never been to but may be in-terested. Traditionally, it centrally collects contextual information and interaction data to model users' preferences, which raises many privacy concerns. The current studies habitually sacrifice the recommendation performance to cope with privacy con-cerns. To protect users' privacy while ensuring the performance of the POI recommendation system, we propose a Ranking-based Federated POI Recommendation with Geographic Effect (RFPG). The RFPG allows users to reserve their private data on local devices to secure privacy. It adaptively constructs an active region to model the geographic effect, enhancing users' personalized preference modeling. In addition, we design a probability-based negative sampling method to protect privacy further and improve recommendation performance. This method calculates the probability of a POI being a negative sample through POI geographic distribution, then combines the positive samples to construct a local triplet training dataset. Theoretical analysis and experiments on two real datasets demonstrate that our proposed RFPG improves the performance of the POI recommendation while protecting users' private data.

As people enjoy the personalized services recommended by Recommender Systems (RSs), the privacy disclosure risk increases with frequent interactions. Malicious adversary often collects public information online to infer private information for illicit profit. As privacy concerns grew, researchers introduced data obfuscation into recommender systems. However, there still exists several limitations in current work. First, although the existing methods effectively reduce the risk of privacy disclosure, they can be detrimental to the quality of the recommendation service. Second, a range of practical issues under the application of recommendation systems are not considered, e.g., long-tail, density, etc. To address those challenges, we propose a novel framework named Want User Defending Inference (WUDI), a high-performance privacy-preserving debiased framework based on data obfuscation. Unlike the original strategies, i.e., adding or removing user ratings, we introduced some novel strategies to generate an obfuscated matrix. Firstly, we define a new method called Cluster Recommend for alleviating the long-tail skewness and data sparsity in RSs. Then we investigate the gender bias in obfuscation and apply a bias mitigating strategy to RSs. Experiments on public datasets demonstrate that WUDI can outperform the state-of-the-art baselines in obfuscation.

The cross-domain sequential recommendation aims to predict the next item based on a sequence of recorded user behaviours in multiple domains. We propose a novel Cross-domain Sequential Recommendation approach with Graph-Collaborative Filtering (CsrGCF) to alleviate the sparsity issue of user-interaction data. Specifically, we design time-aware and relation-aware graph attention mechanisms with collaborative filtering to exploit high-order behaviour patterns of users for promising results in both domains. Time-aware Graph Attention mechanism (TGAT) is designed to learn the inter-domain sequence-level representation of items. Relationship-aware Graph Attention mechanism (RGAT) is proposed to learn collaborative items' and users' feature representations. Moreover, to simultaneously improve the recommendation performance in the two domains, a Cross-domain Feature Bidirectional Transfer module (CFBT) is proposed, transferring user's common sharing features in both domains and retaining user's domain-specific features in a specific domain. Finally, cross-domain and sequential information jointly recommend the next items that users like. We conduct extensive experiments on two real-world datasets that show that CsrGCF outperforms several state-of-the-art baselines in terms of Recall and MRR. These demonstrate the necessity of exploiting high-order behaviour patterns of users for a cross-domain sequential recommendation. Meanwhile, retaining domain-specific features is an important step in the process of cross-domain feature bidirectional transferring.

As the number of online job postings and users grows dramatically, the accuracy and explainability of personjob fit systems are of increasing concern. An explainable personjob fit system can show reasons when making recommendations to both Human Resources and Job Seekers, building trust between uses and recommendation system while providing accurate recommendation results. However, the existing research on content-based person-job fit mainly focuses on 1) dealing with unstructured statements without effectively using structured information in resumes and jobs, and 2) the explanations of the model stay at the level of giving a few sentences, which leads to a lack of explanations. In this paper, we propose an explainable person-job fit model based on the attention mechanism. We model the resume text through a hierarchical attention mechanism and capture the semantic connections between the resume, structured job text, and unstructured job text through a collaborative attention mechanism to better model the job content and provide both structured and unstructured levels of recommendation explanation. Experiments on a large real dataset show that our model outperforms existing baseline models and provides job recommendation reasons at both levels.

In the field of steganography, to effectively hide the secret message, it is of great importance to determine which part of the steganographic cover is suitable for embedding. Currently, most of the existing works focus on the image cover, while few works touch the audio cover case. In this work, we attempt to characterize the complexity of audio for selecting the steganographic cover. Specifically, the original cover is first convoluted with a specially designed adaptive convolution kernel. Based on the residual between the original and the convoluted audio, we derive a quantity for measuring the complexity of each frame for a given audio clip. Experimental results verify the usability of the proposed complexity indicator, suggesting high-complexity audio cover is favorable for data embedding. It is also found that the proposed complexity indicator could further boost the steganographic performance of the state-of the-art audio steganography methods. The source code is publicly available at https://github.com/capzxy/audio-complexity.

Deep learning, especially the convolutional neural network (CNN), has enjoyed significant success in many fields, e.g., image recognition. Recently, CNN has successfully applied to multimedia steganalysis. However, the detection performance is still unsatisfactory. In this work, we propose an improved CNN-based method for audio steganalysis. Specifically, a special convolutional layer is first carefully designed, which could capture the minor steganographic noise. Then, a truncated linear unit is adapted to activate the output of shallow convolutional layer. In addition, we employ the average pooling to minimize the over-fitting risk. Finally, a parameter transfer strategy is adopted, aiming to boost the detection performance for the low embedding-rate cases. The experimental results evaluated on 30,000 audio clips verify the effectiveness of our method for a variety of embedding rates. Compared with the existing CNN-based steganalysis methods, our proposed method could achieve superior performance. To facilitate the reproducible research, the source code will be released at GitHub.

This paper describes a method to extract time-frequency (TF) audio features by tensor-based sparse approximation for sound effects classification. In the proposed method, the observed data is encoded as a higher-order tensor and discriminative features are extracted in spectrotemporal domain. Firstly, audio signals are represented by a joint time-frequency-duration tensor based on sparse approximation; then tensor factorization is applied to calculate feature vectors. The three arrays of the proposed tensor are used to represent frequency, time and duration of transient TF atoms respectively. Experimental results show that exploiting tensor representation allows to characterize distinctive transient TF atoms, yielding an average accuracy improvement of 9.7% and 12.5% compared with matching pursuit (MP) and MFCC features.