An end to end approach for brand recognition in product titles with BI-LSTM-CRF
Mohamed Annis Souames, Larbi Abderrahmane Mohammedi · HAL (Le Centre pour la Communication Scientifique Directe) · 2022
The aim of this paper is to describe an end to end approach using different deep learning architectures to detect brand names in product titles from online stores and online retailers such as Amazon, Ebay, etc. In this paper we developed a named entity recognition model based on a Bi-LSTM architecture with a conditional random field layer using Flair framework, we also explain how the dataset was curated, cleaned and augmented to improve the model performances. Finally, we compare the trained model against few other models trained using the Spacy framework. Our model gave a relatively high f1-score of 0.83 with a good generalisation to real world cases.