Improving the Prospects for Educational Data Mining

Data mining is an important paradigm for educational assessment. The usual assumption is that mining is performed after educational activity with that activity having been designed without regard for the mining process. This paper discusses how the prospects for successful mining can be improved by imposing constraints or biases on the activities and instruments that generate the data. These biases involve one or more of the following: (a) encouraging, requiring or training students to communicate effectively and often during the course of learning activities, (b) building more instrumentation into the learning environment to enable capturing more kinds of data, including evidence of student attention, (c), enriching the logged expressions themselves so that more inferences from them can be made more easily and with general purpose tools, and (d) seeding the log files with reliable assessment data to help anchor subsequent inferences. A variation on the mining paradigm integrates mining methods into the learning environment itself, so that various forms of “articulated assessment” can become practical. Articulated assessment is the coordination of unobtrusive but less reliable assessment techniques with traditional direct-questioning methods in such a way as to follow a policy that balances the needs for accuracy and unobtrusiveness.