Multi-action Recommender

[Pat Ferrel <pa...@occamsmachete.com>]

It’d sure be great if someone was able to try this on their data. Basically we could ingest several actions per user with one primary/recommended action.

For ecom this would be something like (purchase, view-details-page, write-a-review-for, add-to-wishlist, view-category, etc) All of the secondary actions will be analyzed to find the significant cross-action cooccurrences. The item spaces don’t need to be identical and the users for each action will be merged to one user space. The output will be “indicators” tied to the items in the primary action input so it would look something like:

    itemAID, indicators-items-for-A, indicators-items-for-B, indicators-items-for-C, indicators-items-for-Dt, indicators-items-for-E

If this is indexed by a search engine then the query for user-specific reqs is a multi-field query. For user 1 you have history of action A queried against the A indicator field, history of action B queried against the B field, etc. This is actually one query for Solr and super fast.

The result will be a list of items to recommend for an individual—one that takes into account potentially many actions.

The simplest case is using two actions maybe (purchase, view-details-page).

The literature for recommenders has very little to say about using multiple actions so this could be a breakthrough—we’ll only know if we can test it.


On Aug 15, 2014, at 5:03 PM, Ted Dunning <te...@gmail.com> wrote:

On Fri, Aug 15, 2014 at 2:24 PM, Pat Ferrel <pa...@gmail.com> wrote:

>> On Aug 15, 2014, at 9:05 AM, Ted Dunning <te...@gmail.com> wrote:
>> 
>> It is bad practice to use weightings to express different actions.  This
>> may be necessary in an ALS framework, but it is still a bad idea.
>> 
>> A much better approach is to use multi-modal recommendation in which each
>> action is used independently in a cross-recommendation fashion to measure
>> predictive power.  Some thumbs down actions will likely predict some
>> purchases.  If you smash everything together, you won't see that
> subtlety.
> 
> I think that is precisely what I was suggesting with #1. Using thumbs down
> as the secondary action—a cross-cooccurrence indicator. For the [B’A]h_b
> case. Where A is thumbs up and B is thumbs down.
> 
> For that matter about any action that can be recorded for the same user
> set can now be treated as if it has some predictive value for the primary
> action because the cross-cooccurrence indicator will tell you whether that
> is a correct assumption or not.
> 
> If this pans out it seems like a substantially new way of predicting user
> behavior by looking at many correlated actions but recommending only one
> (or few).
> 

Sorry about that.

I didn't recognize what you were saying until I had read it again a few
times.  Unfortunately, I posted first and read after.

You are correct.  This is just what you said.  And it is very exciting new
ground.