Improving english vocabulary learning with a hybrid deep learning model optimized by enhanced search algorithm

Abstract

In this study, we propose a novel deep-learning architecture that is designed to facilitate vocabulary acquisition for second-language learners of English. A hybridized model combining a tuned LSTM and CaffeNet with the EHGS algorithm. The EHGS was selected from the other algorithms including Manta Ray Foraging Optimization (MRFO), Equilibrium Optimizer (EO), Marine Predators Algorithm (MPA), Runge Kutta Optimizer (RUN), and White Shark Optimizer (WSO) since it is the most balanced algorithm out of all of them in terms of exploration vs. exploitation. From a methodological perspective, we adopt a hybrid CNN-based structural approach to enhance the learning of features and the effective processing of temporal information. It uses Oxford English Corpus and WordNet datasets for pre-training to make sure it is robust and effective. The specified model also outperformed very few with comparative evaluations using metrics of accuracy, F1-score, precision, and mean squared error (MSE). Our model showed an accuracy of 0.92 and an F1-score of 0.91 which far surpassed traditional Gaussian and LSTM methods (accuracy of 0.85 and F1-score 0.84). These findings make clear more advanced NLP techniques that can be applied for the development of intelligent education technology that can help non-native English speakers learn new vocabulary at an unprecedented rate. The better results provided by the proposed model mainly reveal its applicability in novel learning environments and offer students personalized, adapted, and immersive learning experiences.