Quality Evaluation of English Blended Teaching using Bidirectional Encoder Representation from Transformers with Fuzzy System

In recent years, evaluation of English teaching plays an important role and has more demand in the digital civilization. It is difficult to evaluate the full course based on exam results or score. To overcome this problem, an effective evaluation system is required that assess the quality of English teaching. Therefore, this research proposes a robust model namely Bidirectional Encoder Representation from Transformers with Fuzzy (BERT-Fuzzy) that builds an effective evaluation system. Initially, data is collected from various university courses which contains both numerical and text data that assists for effective evaluation. Then distinct preprocessing is initiated for text and numerical data through removing stop words, punctuation, special characters and K – Nearest Neighbor (KNN) imputation, min-max scaling, fuzzification respectively. Further, the preprocessed textual data is embedded into BERT for text feature extraction as it effectively captures contextual dependencies between words and results with dense feature representations. Finally, the output of BERT and priorly processed fuzzy numerical score are fused to the decision layer for evaluating quality of blended English teaching. The proposed BERT-fuzzy model acquired greater results with an accuracy of 0.94 when compared to the existing Genetic Algorithm with Back Propagation (GA-BP) model.