After implementing profile tagging and product tagging, you may want to use the various profile tags to create user segments which in turn can be given tailored recommendations. This means we first need to subdivide the total amount of profiles into an unknown number of clusters. The clusters are determined based on overlap in profile tags, using a clustering algorithm.

We generate a human-readable description of each cluster (for example: "users that have expressed interest in fresh craft beers in summer time and dark toasty craft beers in winter time"). This helps us personalize the recommendations, as well as understand each segment. We then use a recommender algorithm to determine a shortlist of products per each product, and then use an LLM to change the order and provide an explanation.

“We think you will enjoy this set of fresh, summer ales, because it's the time of the year again and we know you enjoy citrus IPAs a lot!”

Adding an explanation like this makes the recommendation more personal, which in turns increases trust and consequently the chance of conversion.