Join Challenge CommunityAbout this challenge
Anomaly detection on dynamic knowledge graphs poses a unique and significant challenge in the field of data analysis and knowledge management. This challenge is part of the International Semantic Web Conference (ISWC), bringing together researchers, practitioners, and industry specialists to discuss, advance, and shape the future of semantic technologies. For more information about ISWC, visit their website at https://iswc2024.semanticweb.org/.
Knowledge graphs serve as a cornerstone in various domains, including artificial intelligence, data management, and semantic web technologies. They facilitate efficient data organization by interconnecting entities through meaningful relationships, enabling systems to derive valuable insights and make informed decisions. With their ability to incorporate heterogeneous data from diverse sources, such as databases, text corpora, and sensor streams, knowledge graphs provide a unified view of information, fostering interoperability and data integration.
However, knowledge is not always static
Knowledge evolves over time due to new discoveries, updates, or changes in context. The dynamic nature of knowledge presents opportunities for detecting various contextual shifts or anomalous behaviors. By continuously monitoring and analyzing changes within a knowledge base or graph, it becomes possible to identify emerging trends, evolving patterns, or discrepancies that may indicate shifts in understanding or unexpected events.
This challenge revolves around detecting anomalies within dynamic knowledge graphs
Particularly dynamic knowldge graphs designed for microservices in an AIOps setting. Microservices architecture is central to modern software development, and maintaining a clear understanding of system behavior is crucial. Dynamic knowledge graphs provide a structured representation of microservice relationships but pose challenges in identifying anomalies. Participants aim to develop innovative algorithms to effectively spot and classify anomalies, thus advancing AIOps capabilities and ensuring the reliability and security of microservices-based systems.
Challenge tracks
This challenge concist out of two tracks, participant can either choose one or two tracks.
Historical Dataset
30,534 dynamic knowledge graphs, with 28,516 healthy samples and 2,018 faults distributed among seven different fault types.
Streaming Platform
Participant solutions will listen for incoming KGs, storing and processing information to alert anomalous behaviour promptly.
Classical Evaluation
Evaluation against new testset without labels. Precision and recall evaluation metrics and scores made available on a public leaderboard
Online Evaluation
Evaluation will be performed on new ingested faults and participants' systems have to alert malicious behaviour as fast as possible.
What do we seek?
Participants are encouraged to unleash their creativity without boundaries, fostering solutions that defy convention and embrace originality in its purest form. Some possible research fields are:
Machine Learning
going from structured information from knowledge graphs into dense, low-dimensional vectors, facilitating efficient anomaly representation learning.
Read moreSymbolic AI
blend of symbolic reasoning and neural network techniques, enabling nuanced understanding and detection of irregularities within complex relational data structures.
Read moreStream processing
real-time analysis of streaming data to identify deviations from expected patterns or norms, enabling timely intervention and mitigation of potential issues.
Read moreImportant dates
Two evaluation periods are available, but only the final evaluation period will determine the winners and those eligible to present.
30/3
Call Participation
19/05
Evaluation Round 1
19/05
Start Round 2
19/07
Evaluation Round 2
01/08
Paper Submission Deadline
Organizing committee
Bram Steenwinckel
IDLab, UGent - Imec
Pieter Moens
IDLab, UGent - Imec
Romana Pernisch
Vrije Universiteit Amsterdam, Netherlands
Mehwish Alam
Institut Polytechnique de Paris, France
Pieter Bonte
KU Leuven, Belgium
Sofie Van Hoecke
IDLab, UGent - Imec
Femke Ongenae
IDLab, UGent - ImecAcknowledgements
The challenge is currently supported by:
