Corpus-based methods for the unsupervised grading of short answer questions.

摘要

Information retrieval (IR) approaches to semantic relevancy indexing can be extended from the traditional query-document paradigm to the question-answer paradigm within the context of automatic grading. The focus of this paper is to evaluate the success of unsupervised corpus-based approaches to short answer automatic grading IR, applied the corpus of student responses1 themselves. We illustrate our methods on two datasets, the first a dataset of 270 answers to a quiz question asked in a first semester calculus class at San Diego State University2, and the second a dataset of 29 answers to a quiz question asked in an introductory computer science class at the University of North Texas.