How a trend is calculated

Trending is not the same as popular of all time. The trend score is built on relative growth in
activity: an item whose views have risen fivefold in the last six hours outranks a stable hit with
twice the absolute view count.

A typical formula combines several signals with different weights:

trend_score = w1 * views_delta + w2 * atc_delta + w3 * purchases_delta

where each _delta is the increase over the chosen window compared with the preceding window of
the same length.

Where and why to use it

Trending products answer a specific problem: a new or anonymous visitor has no history, so personal
recommendations are unavailable. Instead of a random set or a static popular list, trending gives a
current read on what the audience cares about right now.

Beyond the homepage, trending works well as a Popular in this category widget — it keeps the
shopper on the listing page and helps surface products they were not deliberately looking for.

Tip: scope trends by category. A shopper in the sports section does not benefit from trending
electronics, however popular those items are site-wide.

Common mistakes

  • A window that is too short — one hour holds too few events for low-traffic categories. The
    result is an unstable list that changes on every page refresh.
  • No novelty filter — permanent hits that have sold steadily for months occupy every slot.
    A threshold is needed: items in the 30-day bestseller top are excluded or deprioritised.
  • One trend block for the whole site — it ignores page context. A category page should show the
    trends of that category, not global ones.