A Joint Knowledge Graph Reasoning Method

Facing the massive data generated by edge intelligent interconnection applications in the mobile edge computing (MEC) environment, timely and efficient data mining has become an urgent technical problem to be solved. Knowledge graph reasoning is a promising solution to the above challenges. However, the traditional knowledge graph reasoning method can not meet the requirements of MEC for low latency and fewer resources. This paper presents a MEC-oriented knowledge graph reasoning method gate recursive unit for logic reasoning (GRULR). Specifically, the technology regards logical rules as variables and trains two models in an iterative manner under the MEC architecture, namely, Rule Miner and Reasoning Evaluator. The two models are deployed in the central cloud and the edge cloud respectively, jointly trained and mutually enhanced. Rule Miner generates rule sequences based on gate recurrent unit (GRU) network, and optimizes network parameters by using high-quality rules generated by Reasoning Evaluator. Experiments show that this method has a good edge reasoning effect, and can generate high-quality logic rules and send them to the central cloud server for sharing.