Harnessing AI for Educational Harmony: A Deep Learning Solution to Identify Abusiveness in Student Evaluations of Teaching

This study addresses the issue of abusive comments in Student Evaluation of Teaching (SET) at the Higher School of Economics (HSE). Despite the rarity of such comments (less than 1%), their impact on educators is significant. Our research aims to develop a natural language processing (NLP) model for accurately categorizing SET comments in Russian as either abusive or non-abusive. We sourced our data from the SET conducted in the second semester of the 2022-2023 academic year at HSE. To create a robust binary classification model, we conducted experiments fine-tuning various models, including OpenAI's embeddings, Multilingual Sentence Transformer, RuBERT, and DistilRuBERT. Notably, OpenAI's embeddings model outperformed the others, achieving the highest F1 score of 79.80%. This result highlights its superior capability in accurately classifying SET comments. The implementation of this NLP model at HSE demonstrates its potential in enhancing the SET process, ensuring a respectful and constructive educational environment by effectively identifying and filtering out abusive feedback.