INFACT is attempting something which is exceedingly difficult --- inferring facets, student models, and student state from very low-level data logs, with limited prior constraints. Nothing we.ve looked at this quarter suggests there are techniques to make these inferences from unstructured, highly detailed log files. Prior structuring information is necessary --- such as an explicitly authored tutorial with an underlying knowledge representation or a represented set of facets with explicit types of evidence.
I.d like to better understand the knowledge representations used in exiting tutorials and tutorial authoring tool kits. Is there any uniformity/standards used for tutorials? Are there exchange formats to represent these? I've found a few interesting ones on the net.
The Bayes stuff was useful for automating assessment. Nothing magical, still need an explicit tutorial and the information for estimating student states, but when this is authored, a lot of machinery becomes available. In fact an initial Bayes Net could come from a simple tutorial outline which is then refined. Good area for work --- techniques/tools to refine the Bayes net as a large number of students work trough a tutorial, either automatic Inference\clustering tools from collected statistics or interactive authoring tools to explore what.s going on.
Bill W blew me away with his insights in extracting information from student logs --- so let's model log analysis processing on Bill.s expertise by treating Bill as an expert at inferring things from log files; create agents from the knowledge elicitation obtained from questioning Bill.