Leveraging Large Language Models for Base Station in Telecommunications

Telecom network issues can disrupt critical services and lead to significant financial losses. This paper focuses on generating solutions for Base Station (BS) problems and prioritizing BS alarms using Large Language Models (LLMs). Our approach enhances telecom network management by reducing downtime and improving emergency reliability. We evaluated six LLMs: Mixtral, Llama, and Gemini for generating solutions; and DistilBERT, DistilGPT-2, and ELECTRA for classifying alarm priorities. The Mixtral model achieved a cosine similarity score of 0.9472, outperforming Llama, Gemini, and traditional models such as Markov in resolving BS issues. Meanwhile, DistilBERT excelled in classifying alarm priorities, surpassing DistiIGPT-2, ELECTRA, and traditional models such as SVM and Naive Bayes regarding accuracy. These findings underscore the potential of LLMs to enhance network stability and reliability within the telecommunications industry, providing significant advantages over conventional models.