The purpose of this paper is to present a framework based on text-mining techniques to support teachers in their tasks of grading texts, compositions, or essays, which form the answers to open-ended questions (OEQ). The approach assumes that OEQ must be used as a learning and evaluation instrument with increasing frequency. Given the time-consuming grading process for those questions, their large-scale use is only possible when computational tools can help the teacher. This work assumes that the grading decision is entirely a teacher’s task responsibility, not the result of an automatic grading process. In this context, the teacher is the author of questions to be included in the tests, administration and results assessment, the entire cycle for this process being noticeably short: a few days at most. An attempt is made to address this problem. The method is entirely exploratory, descriptive and data-driven, the only data assumed as inputs being the texts of essays and compositions created by the students when answering OEQ for a single test on a specific occasion. Typically, the process involves exceedingly small data volumes measured by the power of current home computers, but big data when compared with human capabilities. The general idea is to use software to extract useful features from texts, perform lengthy and complex statistical analyses and present the results to the teacher, who, it is believed, will combine this information with his or her knowledge and experience to make decisions on mark allocation. A generic path model is formulated to represent that specific context and the kind of decisions and tasks a teacher should perform, the estimated results being synthesised using graphic displays. The method is illustrated by analysing three corpora of 126 texts originating in three different real learning contexts, time periods, educational levels and disciplines.
Read full abstract