A validated feedback classification protocol proves durable across representation methods and languages, suggesting that institutional feedback systems don't need to upgrade models to maintain performance—making model choice a practical deployment decision rather than a technical requirement.
This paper tests whether a teaching feedback classification system from 2019 still works well with modern AI models and if it can be used in other languages. Researchers compared the original frozen embeddings against newer transformer models and large language models on Spanish data, then adapted the sentiment classification to English.