Comparison of Artificial Intelligence Models in Cross-lingual Transfer Learning through Sentiment Analysis
Wael Jefry, Firas Al-Doghman, Farookh Khadeer Hussain · 2023
Cross-lingual transfer learning has become an effective way to overcome the limitations of sentiment analysis across various languages. Artificial intelligence (AI) models, particularly those built on transfer learning architectures, have demonstrated remarkable potential in this area in recent years. This paper provides an extensive evaluation of the role played by models in facilitating cross-lingual sentiment analysis through transfer learning techniques. The study provides valuable insights into selecting effective feature representations and extraction techniques for sentiment analysis tasks, enabling researchers to make efficient decisions and improve accuracy and performance in sentiment analysis applications. There are three models, neural network-based (NNB), K-Nearest Neighbor (KNN), and Graph Convolution Network (GCN) models that have been selected and applied over English data to Arabic data after training over other languages. The average accuracy of the Neural Network Based (NNB), K-Nearest Neighbor (KNN), and Graph Convolution Network (GCN) models is 56%, 50%, and 51%, respectively. The accuracy of the NNB model proved the highest, coming in at 56%, while that of the KNN and GCN models was slightly lower, at 50% and 51%, respectively.