On Fri, Aug 28, 2009 at 12:49 AM, tim fulcher
<against.entropy@googlemail.com> wrote:
Hi
Making no bones about it, this is a broad topic request and I've got
little hands-on Drools experience, just looking for some pointers on a
potential Drools approach, feedback appreciated.
Background:
We have a system which asks online users questions to diagnose an
issue (the problem domain isn't important right now).
The questions are structured in a tree/graph such that one question
leads to another until a conclusion - not exactly rocketscience to
this list I suspect! Any question will have n outputs (defined in
configuration); each output is mapped to an option in an HTML
select/radio button.
The key issue though is that this question flow is all running in the
browser, by reading in an XML configuration and parsing this in
javascript. Instead I need this executing server side, i.e. in a rule
engine. One reason is that increasingly questions are being automated
- currently we make an ajax call which decides upon the outcome to
select and therefore move onto the next. This is however a little
clunky at present.
I see Drools as a potential execution engine for our reworked system,
running rules until needing a manual question to be asked, and
triggering question display in a page. Upon the answer the flow
continues in that fashion until a conclusion is reached.
Specific queries:
We have a legacy stack of question flows, how to adapt them into a
syntax amenable to Drools? Can I do this at runtime?
( Our question graphs could be accessed via database access or in an
XML format )
For modeling and processing the questions as fact objects, XML is
well suited. If the XML schema is well designed, you may unmarshal
(using JAXB) to objects which are ready to be inserted as facts.
Given many questions have >2 outcomes, I assume each question-answer
permutation would be a rule?
This would depend on whether distinct activities other than storing the
answer and proceeding to the next question as a function of the current
answer is required. A generic QA handling would be
when
$i : Issue( $a : answers )
$q: Question( $next : next )
Answer( issue == $i, question == $q, $v : $value )
then
mouth.ask( $next.get( $v ) ); // global service
$a.put( $q, $v );
end
Does every question have to be mapped to a bean to become a fact to
operate on? Given our questions are of a few broad types
(boolean,numeric,select one) is that all I need to model?
It's possible that you can have your questions as data in a map, not as
individual facts. The rule engine would then just handle the "abstract
QA algorithm", asking Qs, registering As, pursuing the issue, going
back to previous answers, etc. If this isn't feasible or practical, you
would indeed have all Qs as facts. Presumably a small hierarchy of
classes (AbstractQ, BooleanQ, NumericQ, EnumQ ) is sufficient.
In our current setup a user could decide to re-answer an earlier
question and therefore take an alternate route. What would be the
equivalent in Drools?
Easy: There would be a rule determining that A(Q) is already there;
the fact representing the Q might determine whether this is permitted
etc.
If this is viable, will I need 1 rule outcome to pose a question and
another to operate on the outcome, i.e. will the rules fire up
to/including the decision to ask a question, then when an answer fact
is injected into working memory the rules will fire from that point
onwards?
In such a scenario a session might "run until halt", possibly with one
set of rules being conditioned to fire when there is an unhandled anwer,
and another set prepared to fire when there is no open question in
the WM.
HTH
-W
Thanks
A:E
_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users