Lei Zhang

The influence maximization (IM) problem has attracted the attention of many researchers. However, most of the existing algorithms do not fully use the comprehensive information (local and global information) of the network, which causes the performance of these algorithms to decrease on some networks. To this end, this paper proposes a local-global information guided evolutionary algorithm (LGIEA) for IM problem in social networks. In LGIEA, the local-global information of the network is fully utilized to find the nodes with high potential influence during the evolutionary optimization. Specifically, local information pertains to the neighborhood surrounding the node, whereas global information is associated with the shell value of the node. By leveraging the aforementioned local-global information, we propose an evolutionary algorithm that constructs a novel local-global metric and employs a search space reduction strategy to guide population evolution. Additionally, we design strategies for initialization, crossover and mutation to improve the evolutionary process. Experimental results on 12 real-world networks and four simulated networks verify the effectiveness and efficiency of the proposed LGIEA in comparison with nine representative baseline algorithms and four variants of LGIEA.

Despite recent progress, the Estimation of Model Accuracy (EMA) for protein complexes remains less advanced compared to that for protein monomers. A key challenge lies in effectively integrating both interface‐specific and global structural information to accurately assess the quality of protein complexes. Here, we introduce ContrastQA, the first EMA framework for protein complexes that incorporates the proposed label‐guided graph contrastive learning based on interface quality. By integrating a geometric graph neural network to model global structural features, ContrastQA effectively captures both local (interface‐level) and global (structure‐level) information for accurate model quality estimation. ContrastQA achieved ranking losses of 0.123 and 0.116 on the TMscore and GDT‐TS metrics on the CASP16 dataset, which are 0.015 (10.9%) and 0.012 (8.7%) lower than the second‐best EMA method with ranking losses of 0.138 and 0.128. Our study demonstrates the strong effectiveness of the label‐guided graph contrastive learning module, particularly in selecting high‐quality models. These findings suggest that our graph contrastive learning framework serves as a valuable pre‐training strategy for learning protein structure representations.

Evolutionary algorithms (EAs) have demonstrated excellent performance in many fields, yet their theoretical research remains relatively underdeveloped. As an important combinatorial optimization problem, subset selection provides a critical research avenue to advance the theoretical study of EAs. However, most existing studies focus on solving various variants of subset selection problems, lacking in-depth exploration of more knowledge used in evolutionary operators. This makes it difficult to design EAs that are both more effective and theoretically guaranteed. Therefore, we propose a theoretically guaranteed EA with dual-knowledge integration for subset selection (DKESS). The algorithm integrates two key types of knowledge: the prior fitness contribution knowledge and the population historical statistics knowledge, to comprehensively evaluate the importance of variables and guide the optimization process, thereby enhancing optimization performance with theoretical guarantees. Specifically, we construct a dual-knowledge integration strategy for assessing variable importance, based on which two novel operators are designed: the dual-knowledge guided crossover operator (DKC) and the dual-knowledge guided mutation operator (DKM). These two operators can be used to effectively identify and select high-value variables during the optimization process, thus facilitating the generation of better solutions. Furthermore, we theoretically prove that when the objective function is monotonic, DKESS can achieve the optimal approximation guarantee of 1 − e^−γ within polynomial time. Experimental results on two typical subset selection applications (i.e., sparse regression and unsupervised feature selection) demonstrate that the proposed DKESS outperforms state-of-the-art methods.

Community detection is a fundamental task for understanding complex networks. Many traditional community detection methods only focus on the relationships between nodes in the topology space (i.e., the topology graph) and node attributes. These methods do not fully consider the relationships between nodes in the feature space (i.e., the feature graph) and the global structure of the graph. To this end, we introduce an innovative approach termed Graph Diffusion enhanced Contrastive representation Learning (GDCL) method on topology and feature space for community detection. Specifically, we first construct a feature graph from node attributes to capture the relationships between nodes in the feature space. Based on the topology graph and feature graph, we deploy the graph diffusion module to generate topology diffusion and feature diffusion graphs respectively. Then, we leverage the shared graph convolutional and graph contrastive learning modules to learn node representations of the local and global structures of the graphs. Simultaneously, we employ the hidden layer adaptive embedding fusion mechanism to obtain high-quality node representations. Finally, the embedding integration strategy integrates them with feature embedding to obtain the final embedding for community detection. Extensive experiments on ten real-world datasets demonstrate that GDCL outperforms the existing state-of-the-art algorithms.