Skip to main content Skip to navigation

Automated Assessment of Multimodal Answer Sheets in the STEM domain

Project Overview

The document explores the application of generative AI in education, specifically addressing automated assessment techniques designed for grading multimodal answer sheets in STEM subjects. It identifies the challenges inherent in evaluating assessments that include both textual responses and diagrams, proposing a sophisticated system that leverages Natural Language Processing and Large Language Models to ensure accurate and efficient grading. By minimizing the need for manual intervention, this system aims to enhance the reliability of grading practices, thereby fostering a more equitable educational environment. The findings suggest that the integration of generative AI not only streamlines the assessment process but also improves the overall fairness and effectiveness of educational evaluations, ultimately benefiting both educators and students.

Key Applications

Automated assessment system for grading textual answers and flowchart diagrams in STEM.

Context: STEM education, targeting students in subjects like Data Structures, Computer Networks, and Software Engineering.

Implementation: A two-stage approach involving textual evaluation using LLMs and diagram evaluation utilizing object detection and context-aware analysis.

Outcomes: Improved grading accuracy, efficiency, and fairness; insights into student comprehension; reduced manual grading efforts.

Challenges: Handling the variability in diagrams, maintaining accuracy in OCR for handwritten text, and ensuring contextual understanding in evaluation.

Implementation Barriers

Technological Barrier

Challenges in achieving high accuracy in Optical Character Recognition (OCR) for handwritten text.

Proposed Solutions: Utilizing advanced models like TrOCR and implementing line-by-line segmentation to enhance recognition accuracy.

Evaluation Challenge

Comparing diagrams that may have different representations but convey the same meaning, leading to lower scores based on visual similarity.

Proposed Solutions: Transforming diagrams into textual representations for a nuanced assessment and leveraging LLMs for contextual evaluation.

Project Team

Rajlaxmi Patil

Researcher

Aditya Ashutosh Kulkarni

Researcher

Ruturaj Ghatage

Researcher

Sharvi Endait

Researcher

Geetanjali Kale

Researcher

Raviraj Joshi

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Rajlaxmi Patil, Aditya Ashutosh Kulkarni, Ruturaj Ghatage, Sharvi Endait, Geetanjali Kale, Raviraj Joshi

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies