Where in the Data Stream Are We?: Analyzing the Flow of Text in Dialogue-Based Systems for Learning (ADL Chapter 19 pg 237-245) 12014; Army Research Lab Book); Morrison, D.M.; Nye, B.; Hu, X
This chapter continues a discussion we began in the first volume in this series (Hu, Morrison&Cai, 2013) concerning the use of “learner micromodels” in dialogue-based ITSs. As we originally defined it, a learner micromodel in an ITS is an estimate of a learner’s cognitive and/or affective state at a given time in an ITS session, based entirely on the real-time dynamics of that particular session. An example of such a micromodel, which we call a Learner’s Characteristic Curve (LCC), tracks two features of a learner’s recent dialogue history–novelty and relevance–where novelty is a measure of the degree to which the learner’s contributions to the dialogue add something new, and relevance is a measure of the degree to which his or her contributions conform to an expected answer. This model is called a “characteristic curve” because the learner’s trajectory on these measures over time can be evaluated against certain archetypal curves or boundaries (e.g., thresh-olds). In this chapter, we identify an interesting complication with this design, then expand the discussion to include some general principles concerning the analysis of the data sets and live streams of data that are beginning to flow in vast quantities from Internet-based learning environ-ments, including those with human tutors, artificially intelligent tutors, and, perhaps most interestingly, hybrid systems of the future.