Using Embeddings for Dynamic Diverse Summarisation in Heterogeneous Graph Streams

A high-volume of data generated nowadays by the rise of Smart Cities and Internet of Things can be represented as graph streams. While many graph processing algorithms could analyse small graphs when challenging real-world graphs occur in distributed settings like sensor-based ones, a more suitable analysis is needed. Specifically, challenges like dynamism, heterogeneity, continuity and high-volume of these graph streams could benefit from real-time analysis. This analysis should happen with reduced network traffic and latency while maintaining high data expressibility and usability. Therefore, our key question is: Can we define a dynamic graph stream summarisation system that provides expressive graphs while ensuring high usability and limited resource usage? In this paper, we explore this question and propose a multi-source system with windowing, data fusion, conceptual clustering and top-k scoring that can result in expressive, dynamic graph summaries with limited resources at no expense of usability. Our results show that sending top-k fused diverse summarisation, results in 34% to 90% reduction of forwarded messages and redundancy-awareness with an F-score ranging from 0.57 to 0.88 depending on the k compared to sending all the available information. Also, the summaries' quality follows the agreement of ideal summaries determined by human judges. Nevertheless, these results occur at the expense of higher latency ranging from similar latency to the baseline up to 4 times more depending on the approach; therefore, there is some trade-off between latency, the number of forwarded messages, and expressiveness.