Summarizing Students’ Text-Only Answer Sheet Using SBERT and K-Means Clustering and Evaluating It Using Semantic Search

Covid-19 pandemic has forced the education system to move online. Online classes and examinations have become norm of the day. Reading long article on screen is a tedious and time-consuming task. The current solutions have limitations as a tool for examination in terms of text size, vocabulary, and efficiency in the evaluation. An automatic evaluation system that evaluates students’ text-only answers can reduce the burden of teachers. In his work, we first perform extractive summarization of students’ answer using SBERT and K-means clustering for summary creation. To extract diversity from text, sentences from each cluster are picked based on similarity score. If two sentences have similarity score > 0.8, only one of them is picked for summary generation. The generated summary is evaluated based on results obtained through semantic search based on similarity with the reference summary. The results are promising as the marks obtained by the student using proposed solution is very close to that allotted by teacher after manual evaluation.