#10 Question Selection Strategies: Neural Net & Genetic Algorith
Question selection strategies -- how pavlov chooses
questions -- are pluglets. Source code is in plugsrc in
CVS.
It would be really cool to implement neural net and
genetic algorithm strategies. There's a GPL'd Java GA
package (jaga, i think) at freshmeat that I think could
be made to run with ChapterData as the population and
QuestionData as nodes. Would just have to implement a
couple of methods in the Population and Node
interfaces....
As far as neural-net, I can't quite see what the topology
of the net would be. The only thing that comes to mind
is node A = questiondata A, node B=questiondata B, and
the graph edge represents "answered after." Which
would ferret out when a trick question A throws the user
off and makes him miss question B. The recordkeeping in
pavlov.user doesn't keep this information.
But, somebody else did it. There's a java quiz program
either at sourceforge or freshmeat (probably both) that
chooses questions using a net. Not sure what the
license is.
Logged In: YES
user_id=505143
i've been pondering this a bit... maybe the right model
for a neural net implementation is setting the questions
as nodes, and seeding the initial connection weights with
some metric which compares the similarity of the questions.
this metric could compare the words in the questions (and
maybe the right answers) so that 2 copies of the same
question have a weight of 100% and 2 questions with no
words in common have a weight of 0. Then the net learns
how to make the user get questions wrong....