Enhancing Named Entity Recognition Through Neural Architectures
Wajiha Abdul Shakir · 2024
This study delves into advanced neural architectures for Named Entity Recognition (NER), a task pivotal in Natural Language Processing. Focused on leveraging context dependencies within sentences, our investigation explores the efficacy of incorporating memory using Long Short-Term Memory (LSTM) networks. It extends to Bi-directional LSTMs for capturing both past and future information. Additionally, I examine the integration of Conditional Random Fields (CRF) to model label dependencies. Our findings underscore the substantial performance boost achieved with BiLSTMs, emphasizing their effectiveness in NER tasks. However, the study reveals that introducing CRFs at the prediction step does not significantly enhance predictive quality. Evaluation of the CoNLL-2002 dataset forms the basis for a comprehensive comparative analysis, shedding light on the nuanced role of these architectures in enhancing NER outcomes.