Named Entity Recognition For Macedonian Language

Abstract

Named Entity Recognition (NER), an outstanding technique for information extraction from unstructured texts, is lately becoming the central problem in the field of Natural Language Processing (NLP). In the last few years, multiple Python libraries, like SpaCy, NLTK and FLAIR, accomplished state-of-the-art performances for this problem. As NER is developing into a powerful technique, its real-live applications are becoming more and more numerous: from customer-message categorization to ease of document analysis in greater corporations. In this research, we use a ML-based system with the help of the FLAIR library in Python, which has already provided optimal results for NER in few world-class languages (English, German, Russian, French etc.), for financial entity recognition in financial texts written in Macedonian language. For the NER task on 13 distinct labels using our dataset in Macedonian language on the proposed ML model we have obtained F1-score of around 0.75.