Deep Learning-Based Context-Aware NER for Neologism Detection Across Generations

The rapid proliferation of neologisms in online and social communication has intensified linguistic disparities across generations, posing challenges for timely and accurate neologism detection. Conventional dictionary- and rule-based methods struggle to capture the dynamic, context-dependent nature of evolving expressions, limiting their practical applicability. This study proposes a deep learning-based, context-aware Named Entity Recognition (NER) model designed to effectively identify neologisms and mitigate generational language gaps. The model leverages KoELECTRA, a pre-trained Korean language model optimized for contextual understanding, and is evaluated on large-scale datasets with varying entity distributions. Experimental results demonstrate that the proposed model achieves robust neologism detection performance, with an F1-score of 0.89 under real-world imbalanced conditions and up to 0.92 in controlled, uniform settings. Furthermore, the model exhibits balanced recognition across diverse entity types, confirming its applicability to both formal and informal language environments. These findings highlight the model’s potential as a scalable solution for enhancing linguistic accessibility and supporting real-time adaptation to evolving language, particularly in bridging generational communication gaps.