Dynamic BERT-SVM Hybrid Model for Enhanced Semantic Similarity Evaluation in English Teaching Texts

Abstract

Accurate semantic similarity evaluation in English teaching texts is essential for enhancing automated feedback systems and personalized learning. This study introduces a Dynamic BERT-SVM Hybrid Model, an innovative framework that combines the deep contextual understanding of BERT (Bidirectional Encoder Representations from Transformers) with the robust classification capabilities of Support Vector Machines (SVM). The primary objective is to develop a method that effectively addresses the complexities of language semantics in educational materials by leveraging BERT's ability to generate rich, dynamic embeddings and SVM's proficiency in handling high-dimensional data. The model processes English teaching texts through BERT to obtain nuanced semantic representations, which are subsequently classified by an optimized SVM. Extensive experiments were conducted on a diverse dataset encompassing various genres and proficiency levels. The Dynamic BERT-SVM Hybrid Model outperformed baseline models, including pure BERT and traditional machine learning approaches, achieving higher accuracy, precision, recall, and F1-scores. Additionally, the model demonstrated strong generalizability across different text types, highlighting its adaptability for real-world educational applications. This research bridges advanced natural language processing techniques with educational technology, providing a robust tool for precise semantic evaluation. The Dynamic BERT-SVM Hybrid Model sets a new standard for semantic similarity assessment in language education, offering significant contributions to both academic research and practical instructional methodologies.