Recommendation algorithm

Recommendations are a selection of items on the website and in the mobile app that may be of interest to a particular user. The recommendation system is responsible for selecting relevant items and determining the order in which they are displayed in a separate dedicated section on the website, located below the main item results.

How our recommendation system works:

1. Preparing user data

The system analyses the user’s activity on our website and in the app over a period of up to 180 days:
- Viewing an item card
- Adding an item to favourites
- Adding an item to the basket
- Purchasing an item
Based on these actions, a ranking is calculated for the subcategories to which the relevant items belong. A category ranking is then calculated as the sum of the rankings of all its subcategories.
Minimum thresholds are used for selection:
- if a subcategory’s ranking exceeds the threshold, it is added to the priority list,
- if none of the subcategories within a category reaches the threshold, but the category as a whole exceeds it, selection takes place at category level.

2. Candidate selection

For each relevant subcategory or category, items are selected based on their current availability, the selected time and whether they can be delivered to the relevant city. Subcategories are prioritised — a category is used only when there are no suitable subcategories.
Selection is based on:
- the user’s personal scores,
- popular items in the category,
- popular items in the subcategory.

3. Filtering and updating

Items that are unavailable for purchase or delivery to the selected city are excluded from the final list. Duplicates across different sources are also removed (for example, items selected both through a subcategory and a category).

4. Final ranking

A final relevance score is calculated for the selected items based on:
- subcategory and category rankings,
- the number of interactions with the item (taking into account limits on awarding points for repeated interactions with the same item),
- additional item characteristics, such as price, whether there are reviews, the shop rating and other factors.

5. Displaying items

The system sends a limited number of items to the widgets depending on where they are displayed (usually up to 50 items that users can scroll through). Items are sorted in descending order of their final relevance score. The 18 most relevant items are shown initially, while the rest are loaded when the user clicks ‘Show more’.
Recommendations are generated exclusively using data about user activity collected on the Flowwow website and in the mobile app. Information from third-party websites and services is not used.