Skripsi
CLINICAL NAMED ENTITY RECOGNITION PADA DATA BIOMEDIS MENGGUNAKAN PRE-TRAINED WORD EMBEDDINGS DAN DEEP LEARNING
The rapid growth of digital biomedical data has posed significant challenges in managing and extracting information from unstructured medical texts. This study aims to develop and evaluate a Clinical Named Entity Recognition (CNER) model by combining pre-trained word embeddings with deep learning architectures. Three biomedical datasets were used: JNLPBA, NCBI-Disease, and BC2GM. The experiments were conducted in two stages: the first stage compared the performance of GloVe-BiLSTM, ELMo-BiLSTM, and BERT-BiLSTM combinations; the second stage evaluated BERT-BiLSTM and PubMed2MBERTBiLSTM models using fine-tuning and early stopping strategies. Evaluation using macro average precision, recall, and F1-Score shows that contextual embeddings consistently outperform static embeddings, with GloVe yielding the lowest performance. Transformer-based models like BERT and PubMed2MBERT outperform ELMo due to their self-attention mechanism that better captures token relationships. PubMed2MBERT-BiLSTM, pretrained in the biomedical domain, achieved the best performance across all datasets, highlighting the effectiveness of domain-specific models in medical entity recognition.
Inventory Code | Barcode | Call Number | Location | Status |
---|---|---|---|---|
2507003420 | T175811 | T1758112025 | Central Library (Reference) | Available but not for loan - Not for Loan |
No other version available