GEM: Graph Embedding and Mining


ECML-PKDD 2020 Workshop



Due to the current uncertainty caused by COVID-19 the workshop will take place virtually. Authors will be able to present their work, attend other author/keynote presentations and interact with them using an appropriate virtual conferencing solution.

Overview


Networks of various types, such as property graphs, linked data / semantic web / RDF data, attributed graphs, and more, are increasingly used in practice as versatile and efficient models for data as encountered in today’s data-intensive research and industries. They allow capturing not only the information about entities (using attributes or properties), but also the relational structure between them. This adds significant additional flexibility as compared to data that is formalized as a set of unrelated data points – a data format that was dominant in the machine learning literature until quite recently.

The ability to model and discover knowledge from such network data is therefore fast gaining in importance. Research is needed both into the formalization of new problem types that match well with practical use cases, as well as into the algorithmic, statistical and information theoretic aspects of such problems. The workshop also takes a particular interest in the very popular topic of network embeddings, which as intermediate representation enable learning and mining algorithms devised for non-relational data to be applied to graphs.

Topics of interests include

Unsupervised and representation learning

  • Network embedding
  • Graph compression
  • Entity resolution/deduplication
  • Clustering

Pattern mining

  • Local patterns: community detection, subgraph mining
  • Graph summarization
  • Subgroup discovery on graphs

Supervised learning

  • Classification
  • Link prediction

Exploration

  • Visualization
  • Visual analytics

Modeling

  • Random graph models
  • Opinion formation
  • Information propagation

Data management for graph analytics

  • Data models and structures
  • Index structures

Applications

  • Social network analysis
  • Biological networks and life science data
  • Communication networks
  • Urban data, traffic networks

All types of approaches are welcome, e.g., graph neural networks, traditional ML, random walk methods, Bayesian inference, information theoretical approaches, and we encourage authors to consider the breadth of graph data types (attributed, dynamic graphs, etc.).

The workshop

The workshop will feature:

  • keynote speakers,
  • a few contributed talks,
  • a poster session.