Wed Jan 13 12:58:32 2021

1. Detecting prereq relats among math topics/skills from standard
longitudinal log data


log data: transactions (students' attempts at steps in tutor problems)

usualy interpret them by learning curves on asuccessive application of
same skill

transactons are mostly single skill but sometimes multiskill

AFM models for learning curves

datashop facilities

infer prereq relats from raw transaction data, with help from AFM?

how do prereq relats manifest themselvs in data? (unit or skill
prereqs?  etc.)

typically, evetyone does units in same order (with some tracher driven
variability but not much)

many students and many prolem sets (maybe students over multiple
years)

gaussian correlation networks

(data mostly 0/1 but culd also look at elapsed time, error or hint
counts, etc.  "assistance score = errors + hints")

chatted with Ken K: Ken thought aloud about what measures exacrly
would you be correlating? E;g; estimates slopes per skill/unit, per
student, AFM prob of correctness at end of unit.

other models: PFA (small tweak of AFM for R wizards), BKT, etc.

studnets:  ~ 10000's
skills:    ~ 100's

longitudinal data: less students of course

(longitudinal data may have more real prereq's)


DIAgnostic remidial applications

(building a method might be useful)


2.  summarizing individual students' performance with tutoring
software in a way that is useful for teachers "dashboard"


e.g. incremental update in a BKT model (skill bars)

want to distinguish struggle from successful learning and/or OTL

srtuggle := no change in BKT despite practice

skill bars do not reveal struggle

(seems like agument skill bar (or delta) with # of tiems tried)

same data dimensions


3.  detect small (?) learning discontinuities within tutor log data to
measure effects of out-of-tutor events.  and subsewquent beh


discontinuity in the learning curve at an individual level (change
points in slopes e.g.)

  - bucketing 

etecting cahnge points
  - where we know an interviention occurred
  - where we don't know ...
carnegie learning has unreliable data about when interventions occurred


for teacher feedback on unplanned interventions

Ryan Baker: "the moment of learning" (based on changes in prob in BKT)








----

Carnegie Learning contacts

Steve Ritter

Steve Fonzali