学生反馈对于评估教学质量和教师绩效至关重要，但将文本形式的大量教学反馈自动量化为教师贡献评价指标是一个难题。为此，提出了基于 BERT(bidirectional encoder representation from transformers)和句法依存树的方面级文本情感分析模型，利用教学反馈文本评估与教学质量相关的不同方面，包括师德、教学内容、教学态度、教师能力和学习环境。对于反馈文本采用基于句法依存树的句子嵌入学习，并结合关联词表嵌入，以及基于 BERT 的上下文嵌入，经过多头注意力机制执行特征融合后，提取高质量隐藏特征。其后，使用基于不同机器学习算法的分类器确定情感极性，得到学生对特定教学方面的满意度，从而实现对教师贡献的量化评价。实验结果表明，自由文本形式的学生反馈能够比量表打分更好地衡量不同方面的教学质量。此外，所提框架能够准确提取出反馈文本中不同的教学方面，准确度和 F1 值分别为89。72%和88。91%，性能优于其他方面级情感分析方法。
An automated evaluation system for teacher contributions based on an improved deep learning framework
Student feedback is crucial for assessing teaching quality and teacher performance,but quantifying a large volume of textual teaching feedback into automated evaluative indicators of teacher contributions poses a challenge.To address this,a model for aspect-level sentiment analysis of textual feedback is proposed,utilizing BERT(bidirectional encoder representation from transformers)and syntactic dependency trees.This model evaluates various aspects related to teaching quality in instructional feedback,including moral character,teaching content,teaching attitude,teaching ability,and learning environment.For the feedback text,sentence embedding learning based on syntactic dependency trees is employed,in conjunction with embedding from a lexicon of related words and contextual embedding based on BERT,and perform feature fusion through the multi-head attention mechanism to extract high-quality hidden features.Subsequently,classifiers based on different machine learning algorithms determine sentiment polarity,yielding student satisfaction with specific teaching aspects,thereby achieving a quantitative assessment of teacher contributions.Experimental results indicate that the free-text format of student feedback allows for a better assessment of different aspects of teaching quality compared to rating scales.Furthermore,the proposed framework accurately extracts different instructional aspects from feedback texts,achieving accuracy and F1 scores of 89.72%and 88.91%,respectively,which outperforms other aspect-level sentiment analysis methods.
deep learningsentiment analysisteaching quality evaluationsyntactic dependency treeBERTsupport vector machine
万方数据
维普
