Abstract
Personalized feedback plays a vital role in a student's learning process.While existing systems are adept at providing feedback over MCQ-basedevaluation, this work focuses more on subjective and open-ended questions,which is similar to the problem of Automatic Short Answer Grading (ASAG) withfeedback. Additionally, we introduce the Multimodal Short Answer grading withFeedback (MMSAF) problem over the traditional ASAG feedback problem to addressthe scenario where the student answer and reference answer might containimages. Moreover, we introduce the MMSAF dataset with 2197 data points alongwith an automated framework for generating such data sets. Our evaluations onexisting LLMs over this dataset achieved an overall accuracy of 55\% on Levelof Correctness labels, 75\% on Image Relevance labels and a score of 4.27 outof 5 in correctness level of LLM generated feedback as rated by experts. As perexperts, Pixtral achieved a rating of above 4 out of all metrics, indicatingthat it is more aligned to human judgement, and that it is the best solutionfor assisting students.