Recommendations

Trending

Items rising in views, clicks or purchases over a recent window. The right fallback for anonymous visitors and empty-result pages.

GET /recommendations/trending returns items whose engagement is rising sharpest over the trailing window. Optionally filtered by category, brand, or any indexed facet.

Request

GET /recommendations/trending?indexId=<id>&window=7d&limit=10
Authorization: Bearer <api-key>

Param	Type	Required	Notes
`indexId`	string	yes	Source index.
`window`	string	no	`1h`, `24h`, `7d`, `30d`. Default `24h`.
`limit`	number	no	1–50, default 10.
`filterBy`	string	no	Typesense `filter_by` — e.g. `category:Shoes`.

Response

{
  "items": [
    { "id": "sku-9001", "score": 4.2, "data": { ... } }
  ],
  "window": "24h",
  "engine": "trending"
}

score is velocity — clicks-per-impression ratio over the window divided by the same ratio in the previous window. A score of 4.2 means this item is gaining attention 4.2× faster than its baseline.

When to use

Anonymous homepage — No userId, no anchor item.
Empty results fallback — POST /search returned 0 hits; show trending in the same category. See the no-results recipe.
Editorial widget — "What's hot this week."

Window choice

Window	Best for
`1h`	Live events, breaking news, flash-sale dashboards.
`24h`	News sites, dynamic ecommerce, default for most surfaces.
`7d`	Long-tail stores, B2B catalogues — smooths out daily noise.
`30d`	Bestseller rails — closer to "popular" than to "trending".

Latency

p95 ~40 ms on Pro. The trending bucket is precomputed every 5 minutes per window; API is a hash lookup.

Code path

packages/api/modules/recommendations/procedures/trending.ts.

Frequently bought together

Cart-attach rail driven by co-purchase co-occurrence.

Personalized

Per-user re-ranking driven by analytics history and segment membership. The right engine when you know who the user is.

On this page

Request Response When to use Window choice Latency Code path