Skip to content
Go back

Comparative Evaluation of Pretrained Transfer Learning Models on Automatic Short Answer Grading

Published:  at  02:00 AM

Publication Details

Authors: Sasi Kiran Gaddipati, Deebul Nair, Paul G. Plöger

ArXiv: 2009.01303

Submitted: September 2, 2020

Category: Computation and Language (cs.CL)

Abstract

Automatic Short Answer Grading (ASAG) is the process of grading student answers by computational approaches given a question and the desired answer. Previous works implemented methods of concept mapping, facet mapping, and conventional word embeddings for extracting semantic features, often requiring manual feature engineering.

Approach

We evaluate pretrained embeddings from transfer learning models to assess their efficiency on ASAG:

Unlike previous approaches that extracted multiple features manually, we train using a single feature: cosine similarity extracted directly from the embeddings of these models.

Key Findings

Significance

This work demonstrates that well-chosen pretrained transfer learning models can simplify ASAG by eliminating the need for manual feature engineering while maintaining or improving performance compared to conventional approaches.

Access



Previous Post
Towards AI-Supported Research: a Vision of the TIB AIssistant