Biomedical Information Extraction from Natural Language Processing (NLP) is one of the newest challenges driving innovation in the biomedical scientific field. In this work, we present our implementation pipeline for the GutBrain shared task covering both Named Entity Recognition and Relation Extraction. For Subtask 6.1 (NER), we fine-tuned the GLiNER framework on expert-annotated GutBrain datasets, achieving robust entity recognition between the predefined categories. For the RE Subtasks (6.2.1-6.2.3), we injected entity markers into text and employed fine-tuned BiomedBERT and pubmed-bert classifiers to predict relations between entities. By exploring Precision-oriented, Recall-oriented, and balanced configurations, we identified the best setups for maximizing Precision, Recall, and F1 for each task. Finally, we show our results with scatter plots and discuss the trade-off each run offers.
Named Entity Recognition with GLiNER and Relation Extraction with LLMs
Di Nunzio G. M.
2025
Abstract
Biomedical Information Extraction from Natural Language Processing (NLP) is one of the newest challenges driving innovation in the biomedical scientific field. In this work, we present our implementation pipeline for the GutBrain shared task covering both Named Entity Recognition and Relation Extraction. For Subtask 6.1 (NER), we fine-tuned the GLiNER framework on expert-annotated GutBrain datasets, achieving robust entity recognition between the predefined categories. For the RE Subtasks (6.2.1-6.2.3), we injected entity markers into text and employed fine-tuned BiomedBERT and pubmed-bert classifiers to predict relations between entities. By exploring Precision-oriented, Recall-oriented, and balanced configurations, we identified the best setups for maximizing Precision, Recall, and F1 for each task. Finally, we show our results with scatter plots and discuss the trade-off each run offers.
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
