Label propagation is a promising engine for community detection, widely utilized in anomaly analysis and recommendation systems. However, traditional label propagation algorithms struggle with large-scale networks due to poor stability, slow convergence, and low accuracy. Despite its advancements, the state-of-the-art variant LBLD (Local Balanced Label Diffusion) also faces challenges, such as limited utilization of similarity and a rudimentary approach to constructing the initial community framework, which diminishes the quality of the final results.To address these limitations, we introduce PNLP, a novel similarity-based label propagation algorithm. The key innovations of PNLP include: (i) Integrating similarity throughout the entire process of label propagation, enhancing the robustness and accuracy of community detection. (ii) Defining a Node Importance metric (NI) to provide a more comprehensive assessment of node influence. (iii) Introducing a novel method for identifying rough seed nodes and constructing a tighter initial framework for communities using conductance.Furthermore, PNLP employs a similarity-based influence metric to guide the label updating process, ensuring a more refined and accurate detection of community structures. Extensive experiments conducted on both real-world and synthetic networks demonstrate the superior performance of PNLP, which significantly outperforms established methods such as Infomap, Louvain, Leiden, LPA, and LBLD in terms of accuracy by factors of up to 1.39 times, 1.48 times, 1.44 times, 2.05 times and 1.25 times, respectively.

The influence maximization (IM) problem serve as a promising engine for the comprehension of social multimedia. However, current IM solutions often rely on fixed sampling and diffusion probabilities, which are insufficient to cater to the dynamic needs of online advertising, particularly in targeting users with diverse topic preferences. In addition, existing topic-aware IM solutions for online advertising are either time-consuming during the online query phase or burdened with substantial time and space overhead during the offline index construction phase. To solve the above problems, we propose GraphHash, a topic-aware target influence maximization solution that adopts a combination of offline and online phase. GraphHash constructs offline indexes by utilizing hash-based topic probabilities and selects the seed set with the maximal influence through intelligent indexing technique. In addition, we extend GraphHash using bottom-k sketching technology and apply it to different real-world graphs, achieving significant improvements of orders of magnitude over state-of-the-art baselines.

The graph has recently enabled substantial advances in big data analysis. As graphs are increasing from billions to trillions, efficient graph processing requires large-scale distributed clusters, which have up to thousands of nodes. For big data applications of which the computation is relatively simple, while the communication, especially for imbalanced communication is the bottleneck on distributed clusters, where huge numbers of small messages are transferred through 2D-topology networks. Graph partitioning is the dominant factor to affect the performance of large-scale distributed graph processing. Current graph partitioning policies have paid extensive attention to the utilization of the power law of big graphs but failed to exploit the advanced architectural benefits of 2D topology. To address such a problem, this paper presents GraphMedia, a communication-balanced graph partitioning for distributed search at scale. The key idea of GraphMedia is a communication-balanced partitioning to balance communication based on hardware/software co-design, in which the power law of graphs would be explored to average communication among nodes, and communication would be balanced between row and column by leveraging advanced 2D-topology knowledge. We use both benchmarks and real-world graphs to validate GraphMedia. Specially, GraphMedia-based Graph500 tests on the Tianhe supercomputer are superior to the fastest systems in the latest Graph500 lists (June 2022). We finally apply GraphMedia to real-world graphs for online graph media access, which outperforms the state-of-the-art graph partitioning and graph system by orders of magnitude.