Hello Everyone, I am currently using Typesense to provide product search ability for our e-commerce store. We would like to rank products that were recently purchased higher in the results if the product matches the query. I found the Recommendations cookbook and read through it. Based on my understanding we need to create a new ML model for each user with recently purchased products. Then index this model into a Typesense collection. And then generate embeddings on each search they preform using the original model. I am a bit unsure on how to do this on a per customer basis though because i would like to have each customer have their own unique search instance that is based off of only their purchase history using their customer ID.

Hi Collin, are you looking for recommendations or search personalization? Because recommendations can work on just items the user clicks or views without necessarily involving search queries. E.g. showing similar products.

I am looking for search personalizations on a per customer basis. For each of our customers we would like them to see their recently purchased products rank higher in the results.

We've been working on supporting recommendations in a more plug n play fashion. One problem with using purchase is that you need enough history and purchases for the model to overcome the sparsity problems.

We have over 5 years of purchase history for a majority of our customers from an existing database. I am unsure how to implement the model though. Would we build a new model for each individual customer? or is it possible to tell typesense a list of items(the customers past purchases) to rank higher on each search query.

No you don't have to build a model for each customer. Let me discuss this internally and get back to you.