Automatic Assessment of Short Answer Questions

Automatic grading tools are increasingly used in educational settings as a substitute for manual grading, particularly for open-answer questions. However, many existing systems lack some necessary features to ensure accurate and reliable results and meet the changing needs for different target users. This research addresses some of these limitations by introducing the Automatic Assessment of short Answer Questions system, a novel contribution to the field of Automatic Short Answer Grading. The system utilizes an unsupervised machine learning approach in natural language processing, integrating Word2vec embeddings, the Cosine Similarity metric, and the Ginger spell correction tool to enhance grading accuracy. The inclusion of a spell correction component ensures fair grading for non-native English speakers by mitigating the impact of spelling errors on their marks. Experimental evaluation using the F1 metric demonstrated an average performance of 95%, showcasing the system's effectiveness in improving grading reliability through advanced model configurations and refined spell correction techniques.