The aim of this workshop called Advances in managing and mining large evolving graphs (LEG) is to bring together active scholars and practitioners of dynamic graphs. Graph models and algorithms are ubiquitous of a large number of application domains, ranging from transportation to social networks, semantic web, or data mining. However, many applications require graph models that are time dependent. For example, applications related to urban mobility analysis employ a graph structure of the underlying road network. Indeed, the nature of such networks are spatiotemporal. Therefore, the time a moving object takes to cross a path segment typically depends on the starting instant of time. So, we call time-dependent graphs, the graphs that have this spatiotemporal feature.

Important dates:

  • Paper Submission Deadline: June 25, 2019 (Extended)
  • Author Notification: July 20, 2019
  • Camera Ready Deadline: July 26, 2019
  • Workshop: TBA

Useful links:

First keynote: Danai Koutra, Computer Science and Engineering University of Michigan

Danai is interested in mathematically formulating interesting problems for exploring large-scale networks, and solving them with fast, scalable and principled methods. Her research contributes graph-theoretical ideas and models for exploring a single graph (e.g., graph summarization, inference), as well as multiple graphs simultaneously (e.g., methods for temporal graphs or longitudinal data, similarity and alignment). Real applications of her work include exploration of scientific data, anomaly detection, and re-identification.
Research Interests: large-scale graph mining, mining scientific data, graph similarity, graph matching, graph summarization and visualization, graph anomaly and event detection, data mining, applied machine learning
Title: Efficient Structural Embeddings in Large Time-varying Networks
Abstract: In the past few years, representation learning over networks has been successful in a variety of downstream tasks, such as classification and link prediction. Most existing approaches seek to learn node representations that capture node proximity. In this talk, I will discuss our recent work on a different class of node representations that aim to preserve the structural similarity between the nodes while taking into account temporal information and node attributes. I will focus on an efficient framework, node2bits, which represents multi-dimensional temporal node contexts with (compact) binary embeddings, and then introduce the problem of coupled clustering of time-series and their underlying network. Throughout the talk, I will point out applications to user stitching or entity resolution, entity linking across data sources, and detection of criminal organizations in human-trafficking. I will conclude with challenges and future directions for mining time-varying complex networks.