1
$\begingroup$

When trying to model as a recommendation problem the selection of an item that can be selected (and rated) by the same user many times, I can't find references of previous work.

For example User1 can select item A (and rate it) then item A again (and give it a different rate) and then again item A can be selected n times.. then the same for item B, but item A could be again selected at any time point, and that could happen for any of the A, B, C... Z items. Could this be modelled somehow as a recommender system?

When asked in SO I was told that can't be modelled even as a multi-modal recommneder, then I think my question should be extended to how could I model this for prediction of next rate?

SO question

$\endgroup$
0
1
$\begingroup$

I don't think there is any academic work on the subject, at least that I know of.

One simple way of using that data would be to use the mean of the ratings or other average like measures such as a moving average, a time weighted average, the median, etc.

But this approach is probably not exactly what you're looking for.

Try to look at collaborative filtering approaches with temporal dynamics, there might be something interesting for you.

$\endgroup$
0
$\begingroup$

As to predicting the next rate:

Very simple (kind of brute force) Would be a Dataming approach: Treat it as a timeseries or a lagged model. All or a certain number of past rates are used as inputs to explain the current rate. Input the data into your favorite classifier algorithm (should be something able to handle nominal target variables). Of cause you need a rather large number of observations.

If you prefere a graph approach you might want to take a look into random fields or markov models for sequences.

If this is going to be an online recommendation problem, a graph database would offer the structure and capabilities to efficiently compute the most likely next outcome.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.