Multi-source temporal knowledge graph embedding for edge computing enabled internet of vehicles

Abstract

Edge computing furnishes communicational and computational mediators such as edge nodes (ENs) between vehicles and cloud platforms, accelerating interactive services such as route arrangement and congestion warning for the Internet of Vehicles (IoV). Implemented with the knowledge graph (KG), EN is further enabled to represent the structural relations between multi-source entities (e.g., vehicles and roads) for information reasoning tasks such as traffic flow prediction. However, due to the spatial precision dissimilarity, discontinuity, and relevance of the multi-source data, plus underestimating temporal consecutiveness in traditional rules, establishing KGs for ENs remains a challenge. Given this challenge, a temporal knowledge graph empowered reasoning model named TKGERM is developed. Firstly, confluent data for each EN are initiated by logical smoothing and filtering with multi-source data. Then, the temporal KGs are constructed by extending KG triples into temporal quadruples using timestamp information in the complex vector representation space. Furthermore, with the relative temporal information of past events, logical rules are explored by the temporal KG for traffic information reasoning. Finally, the time performance and effectiveness of TKGERM are evaluated by implementing a series of experiments on real-world data collected from the Bureau of Transportation in Guangdong, China.