LLM Application for Knowledge Extraction from Networking Log Files

In this work, we compare two different techniques, for extracting knowledge from networking log files coming from a LoRA network. Specifically, we compare two different approaches leveraging the few-shot learning capabilities of the LLMs. First, using prompt engineering and few shots, we query an LLM to detect and inspect the correctly received frames in a LoRa network. Subsequently, we ask the same model to produce the Python code to perform the same analysis task on the proposed networking log file, and subsequently we execute the Python code to produce the desired response. While acceptable results are obtained with the first approach, a maximum precision and recall are obtained with the second one, which is actually based on the emerging prompt technique named Chain-Of-Code (CoC).