Entity Recognition in Telecommunications using Domain-adapted Language Models

Large Language Models (LLMs) trained on general data have demonstrated remarkable efficacy in addressing downstream tasks, including Named Entity Recognition (NER). However, technical domains like telecommunications exhibit unique nuances and specialized lexicons that are not sufficiently represented in generic datasets. This study examines the use of encoder language models, pre-trained on an extensive corpus of telecom-specific textual data (TeleText), to enhance Telecom Named Entity Recognition (TeleNER). The contributions of this work are (i) the introduction of two novel Telecom datasets, TeleText and TeleNER; (ii) a comprehensive comparative analysis and evaluation of BERT and TeleBERT models within the TeleNER benchmark; (iii) a reference application of TeleNER for interpreting alarm remedy actions; and (iv) a thorough examination of our findings, accompanied by discussion on potential avenues for future research.