cancel
Showing results for 
Search instead for 
Did you mean: 

filter with rules defined by business user

chholzermsg
Participant
0 Kudos

Hi,

we have the requirement to filter products in product recommendation by criterias, which can be defined by business users (of course they would love a similar functionality like in segmentation).
I first had a look at the pre-filtering in the first (analytical) step. Here I can filter with target groups - which would do the job. But from my understanding this pre-filter is relevant for training the model only, isn't it?
So this leaves the "filter" step. From what I understand, I can register only algorithms defined in HANA there. So this needs a technical guy. In the documentation I have read that HRF is not available for product recommendation. So I would have to implement an algorithm in HANA, which filters e.g. on maritial status and has the status defined as an input parameter, which the business user can select when defining the model, correct? This solution would need an algorithm for each attribute the customer wants to filter on. So it is not flexible and still a very technical solution.
Can you suggest a different way?

Accepted Solutions (1)

Accepted Solutions (1)

angelo_vaikousis
Explorer

Hello Christian,

I would like to first ask you, what exactly is the recommendation scenario?

Do you want to recommend products based on association? i.e. people also looked at these items, people also bought with this product these other products, …

if your criteria can be entered in a target group with the segmentation app then you would be able to recommend products based on on behavior of people that match that criteria. i.e recommend products that other similar users (married or single) looked at or bought together with the product that the user is currently looking at.

For the association based algorithms you are correct. Once pre-filters are added to an algorithm then association rules are created during the generation time for that specific algorithm. This would mean that the algorithm will look only for the products that match the pre-filter. This also means you’re limiting of looking for associations with products that don’t match the pre-filter. That’s okay if that’s what you want to achieve.

Please note that you can define new pre-filters and make them visible in the Manage Recommendations and Recommendation Models apps with our configuration app called Recommendation Data Source Pre-Filters.

Once you use pre-filters you can also pass a specific pre-filter as part of the context from the consuming app. to test the model you can also achieve this when you preview a model. In the Manage Recommendations application once you click the Preview button you can enter parameters (context) that are also used in the model as pre-filters.

You can also create new algorithms. With the Add Custom View app you can import your new HANA views that would be part of an algorithm. You can create new algorithms with our Recommendation Algorithms configuration app by specifying the generation (optional) and runtime HANA views. Please note that you can also export existing HANA views that are included in OOB algorithms and use those as basis of your new algorithms and modify them as needed. Then you can import them in a new algorithm. there’s three type of HANA views that could be imported (Top-N generation, Association generation and Runtime).

Runtime HANA view is mandatory. if you’re creating an algorithm that will be in the Filter step then you only need a runtime HANA view since you will be eliminating product recommendations at runtime from the list of recommendations coming from a previous step.

I hope this helps.

best regards,

Angelo

Answers (1)

Answers (1)

former_member207603
Contributor

Hi Christian,

regarding the first part of your question - the filter settings within the recommendation algorithms/ steps are like positive filters that define

the applicable scope of the alorithm/ step. Meaning it should only be excuted ie. for members of the Target Group (at runtime, not only at design time/ training).

In regards to the second part:

"So I would have to implement an algorithm in HANA, which filters e.g. on maritial status and has the status defined as an input parameter, which the business user can select when defining the model, correct? This solution would need an algorithm for each attribute the customer wants to filter on."

--> correct.

At the moment I don't have another solution for you. However I'll check with a developer to see if there's something I am missing.

Best,

Matthias