What is the difference between "Regular" and "Pre-Computed" recommendations?
Beam offers 3ways to deliver personalization to users; Regular and Precomputed Recommendations.
Live Recommendations are (as you might imagine), live and consistently being updated.
- These are based on user interactions on your website and computed in just a few milliseconds to take into account each click, browse, like, add to cart, checkout, etc as your users are navigating your website.
How it Works
Crossing Minds’ API automatically re-trains the recommendations following event-based triggers (ie User Actions). This means that every time a user provides feedback or intent, the model fine-tunes the recommendations for this user in real-time. This also means that dynamic filtering is always active so, if your users filter out your page based on a tag (color, size, etc) or the price, the recommendation is again adjusted in real-time.
Example
- A user adds items to their cart.
- our API will take into account the add to cart action and signs of preference to provide a fresh set of recommendations based on that action and item properties.
While these are powerful recommendations, they are slightly limited in that they cannot [Do Things (Unsure)]. Simply put there is a trade off in terms of features and quality of recommendations in order to be very flexible
- Can do “session to items”, “user to items with context items” recommendations, …
- As of today, limited by:
- Candidate selection when filters are very restrictive
- No customer “super item / sub item” handling
Precomputed Recommendations are more powerful recommendations that are trained offline.
- This option is best for situations where some time can pass between user action and personalized suggestions. A good example of this would be email campaigns.
How it Works
- Every business rule/filtering/custom handling is permitted
- But everything is set in stone
delivers content to users in seconds
Precomputed or offline recommendations are highly customized but are set in stone after computation.
- Takes time to train the data set.
- These are more powerful recommendations; we take into account [stuff] which makes these recommendations best for email as they are not instant (or live recommendations).
- We retrain the model every 24 hours to include new users, new items, etc
- Much better recommendations and perfect for emails and other recommendations that don’t happen immediately
Multiple options can be selected.
