Exploring methods from other fields: A case study in how Educational Data Mining (EDM) has achieved population validity in the detection of student affect.

Abstract: As linguists look to expand both the breadth of their research subjects and the impact of their work, analysis of larger data sets has become more popular. As we continue to develop as a field, it is important to look at other areas of research which have developed modeling systems for dealing with large, complicated data sets. To that end, this study will present methodology used in Educational Data Mining (EDM), an emerging discipline that leverages large data sets to model student and instructional patterns.

EDM techniques have made it possible to effectively assess a broad range of constructs pertaining to the student, moving from traditional assessment of student knowledge to assessing engagement, affect, strategy, and meta-cognition. In a case study conducted by researchers at Worcester Polytechnic Institute and Columbia Teachers College, we use a standardized field observation protocol to assess students’ affect while they are using educational software. We then synchronize these observations with the log files of the students’ interactions with the software, developing models that can generalize across populations and detect affect in real time.

These detectors work in part, because of cross-validation techniques which emphasize generalizability over exact description. Implications of these procedures, which might complement the explosion of new methodological techniques that sociolinguistic research has seen in the last ten years, will be discussed.