<![CDATA[Named Entity Recognition (NER) is a fundamental task in Natural Language Processing (NLP) aimed at identifying and classifying named entities in text [1], [2]. This study presents a comparative analysis of the Conditional Random Field (CRF) model performance in recognizing biological species entities on two benchmark datasets: Species_800 and Linnaeus. The CRF model was trained using character-based and contextual linguistic features including capitalization, affixes, punctuation, and neighboring context up to two positions. Evaluation was performed using Precision, Recall, and F1-Score metrics via the seqeval library under the BIO tagging scheme. Experimental results show that the CRF model on the Linnaeus dataset achieved an F1-Score of 0.7890 (Precision: 0.9505, Recall: 0.6744), while on Species_800 it achieved an F1-Score of 0.6519 (Precision: 0.7298, Recall: 0.5891). Performance differences are analyzed through dataset characteristics including label distribution, entity density, sentence length, and entity-to-token ratio. Linnaeus yields better overall performance with an F1 gap of 0.1370, attributed to annotation consistency and a lower entity ratio.]]>
Copyrights © 2026
