| Title | Deterrence of Cycles in Temporal Knowledge Graphs |
| Publication Type | Conference Paper |
| Year of Publication | 2022 |
| Authors | Azghandi, Seif |
| Conference Name | 2022 IEEE Aerospace Conference (AERO) |
| Date Published | mar |
| Keywords | Behavioral sciences, Conferences, Data models, data structures, deterrence, Human Behavior, Predictive models, pubcrawl, resilience, Resiliency, Scalability |
| Abstract | Temporal Knowledge Graph Embedding (TKGE) is an extensible (continuous vector space) time-sensitive data structure (tree) and is used to predict future event given historical events. An event consists of current state of a knowledge (subject), and a transition (predicate) that morphs the knowledge to the next state (object). The prediction is accomplished when the historical event data conform to structural model of Temporal Points Processes (TPP), followed by processing it by the behavioral model of Conditional Intensity Function (CIF). The formidable challenge in constructing and maintaining a TKGE is to ensure absence of cycles when historical event data are formed/structured as logical paths. Variations of depth-first search (DFS) are used in constructing TKGE albeit with the challenge of maintaining it as a cycle-free structure. This article presents a simple (tradeoff-based) design that creates and maintains a single-rooted isolated-paths TKGE: ipTKGE. In ipTKGE, isolated-paths have their own (local) roots. The local roots trigger the break down of the traditionally-constructed TKGE into isolated (independent) paths alleviating the necessity for using DFS - or its variational forms. This approach is possible at the expense of subject/objec t and predicates redun-dancies in ipTKGE. Isolated-paths allow for simpler algorithmic detection and avoidance of potential cycles in TKGE. |
| Notes | ISSN: 1095-323X |
| DOI | 10.1109/AERO53065.2022.9843285 |
| Citation Key | azghandi_deterrence_2022 |
Deterrence of Cycles in Temporal Knowledge Graphs