Bayesian networks: A teacher’s view

Bayesian Networks in Educational Assessment

“…12.6 is a useful way to display the results from Bayesian networks. Almond et al (2009a) describe some considerations in the design. First, note that the probabilities are scaled to percentages, as teachers are more used to working with probabilities expressed as percentages than with raw probabilities or fractions.…”

“…There are two reasons for this: first, it means the report is understandable even for individuals with limited color perception, and second, it means that the report is understandable even if it printed or copied using a printer that can only render grayscale images. Almond et al (2009a) also describes some variations on this idea that can be used for reporting at the classroom (or other student group) level.…”

“…Fortunately, averaging the marginal distributions across students provides a good description of the class average. Almond et al (2009a) explores a number of ways of presenting information from Bayesian networks for a class full of students.…”

“…It would be straightforward to implement a teacher's view in which the teacher could monitor a classroom of students (or teams) working through Biomass, and visit the workstations of the students who needed the most help. Almond et al (2009a) describe some possible classroom views, using data from the ACED assessment.…”

Tractable POMDP representations for intelligent tutoring systems

“…Neural networks and very complex decision trees, such as ones that handle many uncertain situations, can be opaque, making decisions that are difficult to explain. In contrast, POMDPs can report their beliefs and recommendations in ways that users who are not expert in Bayesian principles understand [Almond et al 2009]. In summary, POMDPs are capable of modeling learner information that is qualitatively similar to the data successful existing ITSs use.…”