I bet that the name Retrieval-augmented recommendation reminds you of something. That’s right! It’s a similar process to RAG (retrieval-augmented generation) but with a twist. Instead of using a search engine to retrieve the items and then feeding them to a generative AI, you retrieve the items using a search engine and then re-rank them using a recommender (personalization).
This way you get the best of both worlds : any search engine of your choice but with personalization. This is powerful since you can basically build your own personalized search.
Why?
At WPSolr (WordPress search and recommendations plugin), we came up with this new method of search because one of our clients needed a personalized vector (AI) search for their e-commerce website but all the personalized search providers that interested our client were too expensive when including AI. So we decided to make our own.
How does it work?
We decided on the following solution : a Weaviate vector database combined with Recombee recommendations. This way, we could leverage the power of AI using Weaviate vector search and personalize the order (re-ranking) of the results using Recombee trained on the website’s data. We could use the cheapest plan of Recombee while benefiting of vector search.
This new feature has been added to WPSolr and works with Weaviate and Recombee but can also work with anything else. Combine Elasticsearch (keyword search) with Algolia (recommendation) or Apache Solr (keyword search) with Google Retail (personalization).
Try it out for yourself
If you have a WordPress website, you could use our plugin WPSolr. Our enterprise edition offers an easy way of setting up this retrieval-augmented recommendations system (no code and less than an hour to install and configure).
If you are skeptical, trying it out will cost you nothing (WPSolr offers a free trial, Weaviate is a free open-source vector database and Recombee has a free tier).
