This is the second part of the series where I write about Coveo ML model for commerce which offers personalized recommendations to enhance customer engagement. In the first article, we discussed various Coveo ML models, including the Product Recommendation (PR) model and its strategies.
In this blog, prerequisites for PR model is listed below.
Coveo Machine Learning Product Recommendation (PR) models leverage usage analytics (UA) events to get trained with user interactions. This will then recommend products based on the strategies used. Since these are related to products, commerce related UA events are necessary.
Based on the strategies, required events are listed below.
- Strategy: Cart Recommender (cart)
This strategy is used in the cart page where recommendations are made based on the products which are already in the cart. This is based on the UA events from other similar users who either viewed the details page of the product and purchased the product along with the currently added cart products.
UA Events: detail (viewed details page), purchase (completed the purchase of products)
- Strategy: Frequently Bought Together (frequentBought)
This strategy is used in the product details page or even in quick view page where recommendations are made to purchase along with the currently viewing product.
UA Events: purchase (completed the purchase of products). Optional: add/remove product to/from cart.
- Strategy: Frequently Viewed Together (frequentViewed)
This strategy is used in the product details page to recommend other products which are viewed along with the same product by other users.
UA Events: detail (viewed details page). Optional: click (clicking product links from PDF or even from other recommended product lists)
- Strategy: Popular Items Bought (popularBought)
This strategy is used to recommend popular products based on the number of purchases happening at a particular period of time. This can be used in any section of the site where we want to showcase the popular items which are bought by other customers. Usually it is marketed as trending products.
UA Events: purchase (completed the purchase of products)
- Strategy: Popular Items Viewed (popularViewed)
This strategy is used to recommend popular products based on the number of times a product is viewed at a particular period of time. This can be used in any section of the site where we want to showcase the popular items which are viewed by other customers. Usually it is used during a big sale event to showcase the popular items which are being viewed by other customers.
UA Events: detail (viewed details page). Optional: click (clicking product links from PDF or even from other recommended product lists)
- Strategy: User Recommender (user)
This strategy is typically based on the behavior of the user. Recommendations are calculated based on user browsing pattern, product details page views, add/remove or purchase events. This can be used in any section of the site as it is purely based on the user's behavior.
UA Events: detail (viewed details page). Optional: purchase (completed the purchase of products), add/remove to/from cart.
Based on the requirements, appropriate strategy can be used to get the recommended products. Even though we have multiple strategies, we can create one PR model to train the model with all UA events and then use separate searchHub to target a specific strategy and get recommendation based on the strategy.
We will implement this in the next blog.