Behavioural events — views, clicks, add-to-cart, purchases — feed recommendation models with configurable blending and automatic cold-start fallback, so a new product or a new visitor is never left with an empty shelf.
Views, clicks, add-to-cart and purchases — one call, the same session id as search.
Pick a type, an anchor, and optional blend weights — no training pipeline to run.
One endpoint, any anchor — a product, a session, or nothing at all for trending.
similar_itemsNearest neighbours by vector similarity.
bought_togetherMined co-purchase pairs.
related_itemsCo-visitation across views and clicks.
trendingTime-decayed popularity.
sessionBuilt from the current visitor's session.
query_to_itemWhat shoppers clicked after this search.
upsell / cross_sellSame-category or complementary items.
sequentialNext-item transitions mined from real sessions.
graphKnowledge-graph edges (bought-with, requires).
Mix behavioural, vector and graph signal with configurable weights, and fall back automatically to trending or similar-item candidates when a model can't produce enough results — no empty recommendation blocks.
?collection_name=products
&model_id=similar_items
&object_id=42
&per_page=8
// -> ranked candidates, blended with the model's configured weightsExplore more
The best way to evaluate a search platform isn't reading one more comparison table.
Only then make the decision.