Cloud-Edge Collaboration via Preference-based Reinforcement Learning in Semantic Communications

With the development of next-generation mobile communication, semantic communication is considered as a novel technology that focuses on conveying the meaning of a message to the intended users. The key of semantic communication is that the receiver and sender should have the similar background knowledge, hence the semantic knowledge base is the corner-stone of semantic communication. However, how to select an appropriate content as the background knowledge from huge amount of knowledge base data is a challenge in realizing a semantic communication system. Caching partial semantic knowledge bases at edge servers in advance is a promising approach to overcome the above issue. In this paper, we propose a novel semantic knowledge base deployment architecture, in which the specific content of the semantic knowledge base cached at the edge server is determined by the user’s personal preferences, fact relevance, and the sender-receiver matching degree. The semantic knowledge base deployment architecture is affected by conditions such as latency and storage capacity of the edge server. Based on preference-based reinforcement learning (PbRL), we propose a policy ranking algorithm to drive the replacement of the categorical cache for semantic knowledge bases to maximize the hit ratio and the sender-receiver matching degree under the constraints of latency and caching capacity. Simulation results show that our algorithm can obtain higher knowledge base hit ratio and matching degree.