Public English Teaching System based on Large Language Model Meta-Artificial Intelligence

In recent years, the increased usage of English language in various fields has created a need of better English teaching, serving as a common communication medium for people from different linguistic background and cultures. However, the existing Bidirectional Encoder Representations from Transformers (BERT) is limited to learn and capture longterm text formats which is computationally expensive. Therefore, this research proposes Large Language Model Meta Artificial Intelligence (LLaMA) based on transformer structure to understand complex text formats which allows to handle long-term dependencies in text. Initially, the data is collected from English Language Learning-Ensemble Learning (ELL-EL) dataset which consists of different language features. Further, the data is preprocessed with four methods such as stop-word removal, punctuation removal, stemming and lemmatization are used for cleaning and standardizing text features fed into LLaMA model. The proposed LLaMA with transformer-based architecture is used to develop Public English Teaching (PET) system which supports to generate education content based on public queries. Form the results, the proposed LLaMA model achieved outstanding results in terms of accuracy (92.26%), precision (94.53%), recall (89.97%), and F1-score (89.47%) when compared with the existing BERT.