Explaining AI-driven information models for teaching English: combining natural language processing and visualization
Xiufang Zhang, Huaiheng Wu · 2024
The realm of English language pedagogy is experiencing a period of rapid evolution, wherein the emergence of deep learning technology heralds novel avenues and prospects. The endeavor of automating the evaluation of English compositions has garnered escalating interest as a means to address the issues of subjectivity, time inefficiency, and delayed feedback intrinsic to conventional grading practices among English educators, alongside the constraints of limited time and the substantial workload associated with assessing essays in large-scale English examinations. In this study, we advocate for a DCNN-ATT-BGRU framework designed to scrutinize students' compositional structure and facilitate automated scoring within the realm of English instruction, amalgamating bidirectional channel characteristics with multi-tiered textual attributes. Initially, the framework employs a dual-channel Convolutional Neural Network (CNN) architecture to execute comparative feature extraction between the text under evaluation and established standards. Subsequently, it employs Continuous Bag of Words (CBOW) features in conjunction with dual-channel characteristics to effectuate a comprehensive analysis of sentence components. Ultimately, these features are integrated to yield a final automated scoring outcome. Experimental findings evince the framework's capacity to achieve precise classification of words and sentences through multi-tiered feature amalgamation, boasting an average precision of recognition exceeding 85% on publicly available datasets. Furthermore, it demonstrates superior correlation coefficients in scoring performance, thereby offering novel conceptual insights and technical underpinnings for the advancement of AI-driven English instruction.