Personyze Wiki Personyze Wiki docs
Open Personyze
Docs/ Recommendations/ Category & Interest Recommenders — Recommend a Section, Not an Item
Recommendations

Category & Interest Recommenders — Recommend a Section, Not an Item

Two actions that recommend where to go next rather than what to buy — a product category or an article topic — with the eleven algorithms behind them and three ways to lay them…

5 min read Updated 9 hours ago

Every other recommendation widget answers “which item should this visitor see”. These two answer
a different question: which part of the site should they go to next — a product category, or an
article topic.

It is a small difference with a large consequence. A visitor who does not yet know what they want cannot be sold a
specific jacket; they can be shown that Outerwear is where their kind of visitor ends up.

The two actions

Action Recommends Draws on
Category Recommender Product categories The product catalogue and product behaviour — views, carts, purchases
Recommend interests Article topics — your interests The content catalogue and reading behaviour — reads, goals reached

They are the same engine in two vocabularies, and both are built in the same three-step editor as any other
recommendation: Select Recommendation Algorithm, Customize Look & Feel,
Placement, Trigger & Frequency.

Both run on the page and in mobile apps through the SDK.
Unlike the item recommenders they have no email or JSON-feed variant — a category link only means something where
somebody can click it.

When to reach for one

  • The homepage of a first-time visitor. There is no browsing history to recommend products from, but
    there is always a most-popular category — and it is a better first question than a carousel of arbitrary
    items.
  • After a purchase. The thank-you page is the wrong place for another of the thing they just bought,
    and the right place for the aisle next door.
  • A dead end — an empty search, a 404, an out-of-stock product. A category is something to
    offer when there is no item to offer.
  • Publishing. On a content site, “more like this” at the topic level often beats another
    single article, because it commits the reader to a direction rather than one more page.

What it can show

Each recommended category or topic can carry its name, an image, and a
count of what is inside it — “6 items”, “3 articles”.
The count is worth showing: it is the difference between a label and a promise of something behind it.

The designs

The same recommendation, three ways. Which one you pick is a layout decision, not an algorithm one — every
design works with every algorithm below.

Three designs for a product-category recommender: tiles, strip and list
One recommendation, three layouts — tiles, a round strip, and a list with counts. Click to enlarge.
  • Tiles — image-led cards with the name set over the photograph. The most prominent, and the
    one that works as a homepage section in its own right.
  • Strip — round thumbnails in a row, the pattern shoppers already know from category
    navigation. Quiet enough to sit under a header.
  • List — names and counts, no images. For sidebars, narrow columns, and catalogues whose
    categories have no artwork worth showing.

The interest recommender uses the same three, in the content vocabulary:

The same three designs recommending article topics
The interest recommender, same shapes, content vocabulary. Click to enlarge.

The algorithms

Eleven, and they pair across the two actions — the same engine, named for what it reads. Pick the row, then
the column for the kind of site you have.

Product categories Article topics What it picks
Personalized Recommendations Personalized Recommendations What visitors with similar demographics engage with. The best homepage default.
Most Popular Category Most Popular Interest Site-wide, by views.
Most Bought Category Most Converting Interest Site-wide, by purchases — or by goals reached.
Recently or Most Viewed Category Recently or Most Read Interest This visitor’s own history: where they have actually been.
Category of Recently Viewed Product Interest of Recently Read Article Inferred upward from the individual items they looked at.
Others Who Viewed Also Viewed Others Who Read Also Read A co-view collaborative filter, one level up from items.
What others Bought, who Viewed this What Converting Readers Were Interested In A view→buy, or read→goal, collaborative filter.
You May Like You May Like Personalized, but anchored to the product or article on the page.
Currently Viewing Currently Reading The category or topic of the item on this page, when the page reports one.
Past Orders Recommendations Past Goals Recommendations Related to what this visitor bought, or the goals they reached.
Buy it Again Reached Goal Again Categories they bought from, or topics they converted on, before.
The page you choose still changes the anchor. As with every recommendation, an algorithm anchored to
“the current product” means something different on a Product page than on Home. Pick the page first —
see Recommendation algorithms.

What it needs before it can return anything

  • For categories: a product catalogue whose items carry categories, and category tracking on your
    pages so the engine knows which category a visitor is in. See
    Categories
    tracking
    .
  • For topics: interests
    set on your articles — from the feed, from the contextual engine, or by hand.
  • An image per category, if you want the tile or strip design. The list design needs none.
If it has nothing to recommend, it renders nothing at all — no empty box, no heading with a gap
under it. That is deliberate, but it means “I do not see my widget” and “the algorithm came back
empty” look identical on the page. The QA bar
tells the two apart: it names the algorithm and what that algorithm needs.

Related

Did this page answer your question?
Thank you — that goes to whoever maintains this page.