The core issue in ridesharing is designing reasonable algorithms to match drivers and passengers. The ridesharing matching problem, influenced by various constraints such as weather, traffic, and supply–demand dynamics in real-world scenarios, requires optimization of multiple objectives like total platform revenue and passenger waiting time. Due to its complexity in terms of constraints and optimization goals, the ridesharing matching problem becomes a central issue in the field of mobile transportation. However, the existing research lacks exploration into the fairness of driver income, and some algorithms are not practically applicable in the industrial context. To address these shortcomings, we have developed a fairness-oriented dynamic matching algorithm for ridesharing, effectively optimizing overall platform efficiency (expected total driver income) and income fairness among drivers (entropy of weighted amortization fairness information between drivers). Firstly, we introduced a temporal dependency of matching outcomes on subsequent matches in the scenario setup and used reinforcement learning to predict these temporal dependencies, overcoming the limitation of traditional matching algorithms that rely solely on historical data and current circumstances for order allocation. Then, we implemented a series of optimization solutions, including the introduction of a time window matching model, pruning operations, and metric representation adjustments, to enhance the algorithm’s adaptability and scalability for large datasets. These solutions also ensure the algorithm’s efficiency. Finally, experiments conducted on real datasets demonstrate that our fairness-oriented algorithm based on reinforcement learning achieves improvements of 81.4%, 28.5%, and 79.7% over traditional algorithms in terms of fairness, platform utility, and matching efficiency, respectively.

Drug–drug interactions (DDIs) are entities composed of different chemical substructures (functional groups). In existing methods that predict drug–drug interactions based on the usage of substructures, each node is perceived as the epicenter of a sub-pattern, and adjacent nodes eventually become centers of similar substructures, resulting in redundancy. Furthermore, the significant differences in structure and properties among compounds can lead to unrelated pairings, making it difficult to integrate information. This heterogeneity negatively affects the prediction results. In response to these challenges, we propose a drug–drug interaction prediction method based on substructure signature learning (DDI-SSL). This method extracts useful information from local subgraphs surrounding drugs and effectively utilizes substructures to assist in predicting drug side effects. Additionally, a deep clustering algorithm is used to aggregate similar substructures, allowing any individual subgraph to be reconstructed using this set of global signatures. Furthermore, we developed a layer-independent collaborative attention mechanism to model the mutual influence between drugs, generating signal strength scores for each class of drugs to mitigate noise caused by heterogeneity. Finally, we evaluated DDI-SSL on a comprehensive dataset and demonstrated improved performance in DDI prediction compared to state-of-the-art methods.

In real-world network scenarios, modal absence may be caused by various factors, such as sensor damage, data corruption, and human errors in recording. Effectively integrating multimodal missing data still poses significant challenges. Different combinations of missing modes can form feature sets of inconsistent dimensions and quantities. Additionally, effectively merging multimodal data requires a thorough understanding of specific modal information and intermodal interactions. The abundance of missing data can significantly reduce the sample set size, leading to learning interaction features from only a few samples. Moreover, there is a lack of clear correspondence between heterogeneous data from different sources. To address these issues, we focus our research on multimodal knowledge graph scenarios with different types of structures and content and develop a new knowledge graph embedding method. First, we use three embedding components to automatically extract feature vector representations of items from the structural content, textual content, and visual content of the knowledge graph. Then, we divide the dataset into several modal groups and model these modal groups using a multilayer network structure, with each multilayer network corresponding to a specific multimodal combination. Subsequently, we construct corresponding multilayer network projection layers and propose a two-stage GAT-based transfer learning framework for the projection layers, in which the extracted incomplete multimodal information and intermodal interaction information are integrated and mapped to a low-dimensional space. Finally, we not only theoretically prove the feasibility of the proposed method but also validate its effectiveness through extensive comparative experiments on multiple datasets.

Heterogeneous data and models pose critical challenges for federated learning. However, the traditional federated learning framework, which trains the global model by transferring model parameters, has major limitations; it requires that all participants have the same training model architectures, and the trained global model does not guarantee accurate projections for participants’ personal data. To solve this problem, we propose a new federal framework named personalized federated learning with semisupervised distillation (pFedSD), which ensures the privacy of the participants’ model architectures and improves the communication efficiency by transmitting the model’s predicted class distribution rather than model parameters. First, the server adopts the adaptive aggregation method to reduce the weight of low-quality model predictions for the model’s predicted class distributions uploaded by all clients, which helps to improve the quality of the aggregation of the prediction class distribution. Then, the server sends it back to the clients for local training to obtain the personalized model. We finally conducted experiments on different datasets (MNIST, FMNIST, and CIFAR10), and the results show that the model performance of pFedSD exceeds the latest federated distillation algorithms.