Διπλωματική εργασία--Πανεπιστήμιο Μακεδονίας, Θεσσαλονίκη, 2024.
(EL)
The application of machine learning methodologies in networks requires an appropriate representation of network data in vector form. Several embedding methods already facilitate the representation of network information at graph, node, and community levels. However, the majority of research focuses on static graph snapshots, largely ignoring any temporal network dynamics. As a result, the outcome of network analysis tasks -such as graph visualisation or community detection- is informed by overly reduced data. In response, this work attempts to model temporal dependencies in graphs by introducing ComE+, a dynamic graph embedding and community detection framework which extends the standard ComE clustering algorithm by employing CTDNE’s temporal embedding approach. The proposed model is tested in a variety of datasets compared to several established baselines, proving its capacity for yielding more meaningful network community estimates by relying on time-sensitive node representations.
(EN)
Submitted by ΑΡΗΣ-ΙΑΝΟΣ ΓΕΜΕΝΕΤΖΗΣ (
[email protected]) on 2024-10-22T08:20:40Z
No. of bitstreams: 2
license_rdf: 1031 bytes, checksum: 934f4ca17e109e0a05eaeaba504d7ce4 (MD5)
GemenetzisArisMsc2024.pdf: 2009312 bytes, checksum: b6fa771f5f4d92e8f27b769fdfe82860 (MD5)
(EN)
Approved for entry into archive by Κυριακή Μπαλτά (
[email protected]) on 2024-10-23T09:46:14Z (GMT) No. of bitstreams: 2
license_rdf: 1031 bytes, checksum: 934f4ca17e109e0a05eaeaba504d7ce4 (MD5)
GemenetzisArisMsc2024.pdf: 2009312 bytes, checksum: b6fa771f5f4d92e8f27b769fdfe82860 (MD5)
(EN)
Made available in DSpace on 2024-10-23T09:46:14Z (GMT). No. of bitstreams: 2
license_rdf: 1031 bytes, checksum: 934f4ca17e109e0a05eaeaba504d7ce4 (MD5)
GemenetzisArisMsc2024.pdf: 2009312 bytes, checksum: b6fa771f5f4d92e8f27b769fdfe82860 (MD5)
Previous issue date: 2024
(EN)
University of Macedonia
