Hello,
Today we will share how videos are recommended to you on product pages, for transparency sake.
Explaining tags
First, you have to know that each product has tags associated to it.
Tags can be model names, situations, parts of the body tickled, etc …
The first row (and more accurate) works like this :
– Only videos with at least 1 common tag can be shown as a recommandation, this is a broad filter but it helps keeping things relevant
– It produces a list of thousands of results, but we will only show 4 of them to the user
So how do we decide which ones to show ?
- First, if you were on a clip page in the last 48 hours, we will not show these clips
- Then we have an algorithm that will sort the result.
- Clip Freshness (weighs 1 out 5)
- Randomness (weighs 1 out 5)
- Others that bought this clip also bought X (weighs 4 out 5)
- Clips bought together (same order) (weighs 5 out 5)
- Conversion Rate of the clip (weighs 1 out 5)
Then we have a second rank of recommandation that is only based on common tags, it is broader and is more like a backup.
So why it works ?
The whole goal is to give visibility to clips that you will like.
If we show random clips, and you have specific taste, it will be a waste of Kb on the page.
But we don’t want to use customer data, there is no profiling of any sort, or machine learning
Everything is based on the basic principle that :
- You buy what you like
- You have roughly the same tastes of others
- And of course serendipity is maintained.