Yizhou Li’s research while affiliated with Sichuan University and other places

Social media bots pose potential threats to the online environment, and the continuously evolving anti-detection technologies require bot detection methods to be more reliable and general. Current detection methods encounter challenges, including limited generalization ability, susceptibility to evasion in traditional feature engineering, and insufficient exploration of user relationships. To tackle these challenges, this paper proposes MRLBot, a social media bot detection framework based on unsupervised representation learning. We design a behavior representation learning model that utilizes Transformer and a CNN encoder–decoder to simultaneously extract global and local features from behavioral information. Furthermore, a network representation learning model is proposed that introduces intra- and outer-community-oriented random walks to learn structural features and community connections from the relationship graph. Finally, the behavioral representation and relationship representation learning models are combined to generate fused representations for bot detection. The experimental results of four publicly available social network datasets demonstrate that the proposed method has certain advantages over state-of-the-art detection methods in this field.

With the widespread use of computers, the amount of malware has increased exponentially. Since dynamic detection is costly in both time and resources, most existing malware detection methods are based on static features. However, existing static methods mainly rely on single feature types of malware, while few pay attention to multi-feature fusion. This paper presents a novel multi-feature extraction and fusion method to effectively detect malware variants by combining binary and opcode features. We propose a stacked convolutional network to capture the temporal and discontinuity information in the function call of the binary file from malware. Additionally, we adopt the triangular attention algorithm to extract code-level features from assembly code. Additionally, these two extracted features are aligned and fused by the cross-attention, which could provide a stable feature representation. We evaluate our method on two different datasets. It achieves an accuracy of 0.9954 on the Kaggle Malware Classification dataset and an accuracy of 0.9544 on a large real-world dataset. To optimize our detection model, we conduct in-depth discussions on different feature extractors and multi-feature fusion strategies. Moreover, a visualized attention module in our model is provided to explain its superiority in the opcode feature extraction. An experimental analysis is performed against five baseline deep learning models and five state-of-the-art malware detection models, which reveals that our strategy outperforms competing approaches in all evaluation circumstances.

Breast cancer is a heterogeneous disease and can be divided into several subtypes with unique prognostic and molecular characteristics. Classification of breast cancer subtypes plays an important role in the precision treatment and prognosis of breast cancer. Benefitting from the relation-aware ability of graph convolution network (GCN), we presented a multi-omics integrative method, attention-based GCN, for breast cancer molecular subtype classification using mRNA expression, copy number variation, and DNA methylation multi-omics data. Several attention mechanisms were performed and all exhibited effectiveness in integrating heterogeneous data. Column-wise attention-based GCN outperformed the other baseline methods, achieving AUC of 0.9816, ACC of 0.8743 and MCC of 0.8151 in 5-fold cross validation. Besides, Layer-wise Relevance Propagation (LRP) algorithm was used for interpretation of model decision and could identify patient-specific important biomarkers which were reported to be related to the occurrence and development of breast cancer. Our results highlighted the effectiveness of GCN and attention mechanisms in multi-omics integrative analysis and implement of LRP algorithm could provide biologically reasonable insights into model decision. Author Summary Identification of molecular subtype is essential to understanding the pathogenesis of cancer and advancing the development of cancer precision treatment. We have developed a graph convolution network architecture to predict the molecular subtype of breast cancer patients based on multi-omics data. The major difference between our work and other methods is that we have offered a new view for multi-omics data integration with the assistance of graph convolution network. Besides, we adopted an interpretability method to explain the model decision and prioritize the genes for biomarker discovery.