Recommend for small data set + short lived products


I use Algolia for used car dealership vehicle searching and filtering. I’d like to use the new Recommend product but cannot figure out if it works for this application.

Specifically, used vehicles are a small data set relative to the demo applications (20-60 vehicles in the data set at a time) and those products are constantly moving in and out of the set.

Obviously the vehicles share attributes that can/should power Recommend, like people who view trucks want to see more trucks.

Is Recommend appropriate for that kind of small, ephemeral dataset? Is it smart enough to abstract the common attributes and recommend based on that?


Hi there,

No, currently recommend is designed for a large number of transactions per product (1k - 10k events, depending on the model). It would be challenging to build up enough transactions to train the models, but not impossible. You would need an index that “buckets” the records in some way (“Sedans” or “Hondas”) and then feed your click events to this abstracted index along with your main index. This may allow you to collect enough click events to train the models across the 90 day window.

1 Like

Great answer, thank you Chuck!

1 Like