In this project, we want to implement a scalable distributed system for Massive historicAl Graph MAnagement (tMAGMA) that will be able to store efficiently temporal graphs and in addition provide a query engine with a set of simple and complex standard operations. Additionally, we conduct research on advanced analytics involving local community detection algorithms on temporal networks with seeding sets of particular characteristics. Moreover, we target to find outliers in such a time-evolving network that correspond to nodes or groups of nodes (e.g., agents or communities) that exhibit "peculiar" behavior as the network evolves. These advanced analytics tools will be implemented on tMAGMA. Our vision is to provide a system  that will be the tool of choice for management of time-evolving networks with a library of advanced data analysis tools.

Target Outcomes:

1. Novel network algorithms: our main goal is to design distributed algorithms for community detection and outlier detection in temporal networks. Our first step is to look at centralized semi-streaming algorithms (insertions only) focusing mainly on community detection. Then we will extend it to a full streaming setting and if possible further extend it to allow for valid time. This means, that each change will carry its valid interval with it. Outlier detection will be based on the community detection results. Then, a distributed framework for community detection will be considered based on the "Thinking Like a Vertex" paradigm. We mainly consider local community detection.
2. Α system prototype: we aim to develop a system prototype for managing and querying historical graphs that continuously change. To accomplish this we will use an open-source existing system for managing static graphs (Janusgraph) and change it accordingly.
3. Thorough evaluation: the effectiveness of the system will be verified through extensive experimentation, using established benchmarks modified to accommodate historical information, such as the LDBC benchmark. We will also evaluate the algorithms developed for community detection and outlier detection.

Publications

1. State-of-the-art in Community Detection in Temporal Networks. K. Christopoulos and K. Tsichlas. In Proc. of the 18th International Conference on Artificial Intelligence, Applications, and Innovations (AIAI) - Mining Humanistic Data Workshop (MHDW), 2022.

2.  MAGMA: Proposing a Massive Historical Graph Management System. A. Spitalas and K. Tsichlas. In Proc. of the 7th International Symposium on Algorithmic Aspects of Cloud Computing ALGOCLOUD, pp. 42-57, 2022.

 3. Dynamic Local Community Detection with Anchors. G. Baltsou, K. Christopoulos and K. Tsichlas. In Proc. of the 11th Intern. Conference on Complex Networks and their Applications, 2022.

4. Local Community Detection: A survey. G. Baltsou, K. Christopoulos and K. Tsichlas. IEEE ACCESS, 2022.

5. Local Community Detection in Graph Streams with Anchors. K. Christopoulos, G. Baltsou and K. Tsichlas. Information, 14(6): 332, 2023.

6. Adopting Different Strategies for Imporving Local Community Detection: A Comparative Study. K. Christopoulos and K. Tsichlas. In Proc. of the 12th Int. Conference on Complex Networks and their Applciations, 2023.

7. Local Community-Based Anomaly Detection in Graph Streams. K. Christopoulos and K. Tsichlas. In Proc. of the 20th Int. Conference on Artificial Intelligence, Applications and Innovations, 2024

8. Degree Distribution Optimization in Historical Graphs. A. Spitalas, C. Kapeletiotis and K. Tsichlas. Presented in the 9th Int. Symposium on Algorithmic Aspects of Cloud Computing (ALGOCLOUD), 2024 - Best Paper Award.

9. Triangle Counting in Large Historical Graphs. K. Christopoulos, E. Daskalakis, A. Bompotas and K. Tsichlas. To be presented in the 40th Annual ACM Symposium on Applied Computing (SAC), 2025.

10. Degree Distribution Optimization in Historical Graphs. A. Spitalas, C. Kapeletiotis and K. Tsichlas. To be presented in the 40th Annual ACM Symposium on Applied Computing (SAC), 2025.

Deliverables