Leveraging LLM for Evolving and Declarative Trace Analytics towards Next Generation Mobile Core Networks
Guillermo Rodríguez-Navas, Dongjin Lee, Kris Kim · 2024
The recent developments of the telecom infrastructure (the so-called 5G and 6G networks) lead us to a situation in which diverse networks will coexist, providing services that may evolve over time. Therefore, analytics intended to this context must be flexible enough to handle evolving communication semantics; any method based on a rigid definition of semantics will soon be outdated or become, at best, incomplete. In this paper we present a framework for definition and verification of analytics that is generation-agnostic. This framework leverages a Large-Language Model (LLM) to provide identification of events and network procedures and uses this information to generate suitable analytics on-the-fly. We apply our framework to implement a service for detection of abnormal UE and network behavior and compare it with the mechanisms proposed by the Network Data Analytics Function (NWDAF), which is the current state of the art.