Research Article
Comparative Study of Name Entity Recognition Models in Burkina Faso Context
@INPROCEEDINGS{10.1007/978-3-031-81573-7_10, author={Sibiri Tiemounou and Wend Yam Serge Boris Ou\^{e}draogo and Moumouni Djibo and Yaya Traor\^{e} and Ali Ma\~{n}ga and Souleymane Zio and Fran\`{e}ois Zougmor\^{e}}, title={Comparative Study of Name Entity Recognition Models in Burkina Faso Context}, proceedings={Towards new e-Infrastructure and e-Services for Developing Countries. 15th International Conference, AFRICOMM 2023, Bobo-Dioulasso, Burkina Faso, November 23--25, 2023, Proceedings, Part II}, proceedings_a={AFRICOMM PART 2}, year={2025}, month={2}, keywords={Name Entity Recognition Natural language processing Neural Network NER models}, doi={10.1007/978-3-031-81573-7_10} }- Sibiri Tiemounou
Wend Yam Serge Boris Ouédraogo
Moumouni Djibo
Yaya Traoré
Ali Maïga
Souleymane Zio
François Zougmoré
Year: 2025
Comparative Study of Name Entity Recognition Models in Burkina Faso Context
AFRICOMM PART 2
Springer
DOI: 10.1007/978-3-031-81573-7_10
Abstract
Name Entity Recognition (NER) is an important core component of Natural Language Processing (NLP) systems for identifying entities like person names, locations, and organizations. Many NER models have been proposed in the literature those whose architecture are based on deep neural networks are the most efficient. Burkina Faso, a West African country, is a French-speaking country with its culture and its specificities in the use of the French language. Our work consists in building a user-friendly NER model that reliably identifies 8 entity types from Burkina Faso media information where the French language is the most spoken. To achieve this goal, we firstly build a dataset that has been labeled along these entity types. Then, in order to choose the best architecture, we assessed 6 multilanguage NER models namely Spacy (with its three pre-trained models), Flair, Stanza, and Camembert NER. This paper presents the performance evaluation of existing French NER models when applying to media news data of Burkina Faso. They have been assessed along 3 common entity types Person (PER), LOCation (LOC), and ORGanization (ORG). Results show that Stanza and Flair outperform all models under study with a percentage greater than 70%. They reliably identify a person’s name and location entities. However, their performance is relatively fair in correctly extracting organization entities due to the Burkina Faso context.
