Beyond Automated Scoring: Process-Level Feedback for Conceptual Computer Networking Answers with a Locally Deployable LLM
Year: 2026
Type: conference-paper
Abstract: Formative feedback is a bottleneck in computing education: at scale, instructors often return only a numeric score on short-answer questions, while existing LLM options are either privacy-sensitive commercial APIs or open models giving unstructured feedback. We fine-tune Mistral-7B for fully local deployment and guide it with a Chain-of-Thought prompt that produces process-level feedback—strengths, weaknesses, and next steps. On the SAF Communication Networks dataset, scoring consistency matches zero-shot GPT-4, and in a blinded study 21 senior CS students rate the feedback significantly higher in clarity, specificity, and usefulness (p < 0.01) than a deployable baseline.
Authors: Xixiang Zhang, Ayşe Saliha Sunar
Institution: University of Warwick
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: hybrid
OpenAlex ID: W7204173101