A Comprehensive Evaluation of Machine Learning Classifiers in Named Entity Recognition

Named Entity Recognition (NER) is crucial for organizing unstructured text data by detecting essential features such as names, locations, and organizations. This study provides a comprehensive evaluation of four machine learning classifiers: Conditional Random Fields (CRF), Logistic Regression (LR), Naive Bayes (NB), and Support Vector Machines (SVM). The performance of these models is assessed using the CoNLL-2003 dataset, which is widely recognized for its extensive coverage and intricate structure. The evaluation is based on precision, recall, and overall F1 scores. The results highlight the exceptional performance of the SVM model, which achieved the best F1 score of 89.91%, demonstrating its flexibility in accurately detecting entities. LR showed high effectiveness, especially when there were obvious differences between classes, as indicated by an F1 score of 87.82%. Although CRF and NB had lower performance scores, they showed clear advantages in situations requiring contextual analysis of interaction between classes. This work provides valuable insights for selecting the appropriate classifier for NER tasks to improve the accuracy and efficiency of information retrieval systems.