Recommendations

How does Smart Recommender Work?

Smart Recommender's main focus is to show personalized and relevant content to users. Creating only one strategy for all users is not enough as users are getting more diversified day by day.

To make Smart Recommender work, you need to train the data about what users do and what they are interested in. This data is collected from desktop and mobile web which can also be called clickstream data via System Rules and is processed to be used on recommendations on the Web, Email, App, InStory, and Web Push. Additionally, you should also collect the product catalog of the stores in order to generate recommended product lists for each algorithm. You can do it in 3 different ways; Clickstream data, Catalog Integration through XML, and Catalog Integration through API.

• In Generic Algorithms, product recommendations are based on product performance rather than individual consumer behavior or current context.
• In Contextual Algorithms, product recommendations are based on the current contexts of individual consumers such as the product category of the current page, but not based on consumer behavior or activity.
• As for Collaborative Filtering, personalized approaches seek to identify the right products to recommend content based on what similar users have previously engaged taking product visits, purchases and items added to cart into account.
• The User-Based algorithm works on a user-product-rating matrix. For each product visited by a user, a rating is calculated based on number of visits, purchases and items added to cart within the last 30 days. To recommend products to a user, the algorithm finds similar users to the current user and recommends products from similar users, which are not visited by the current user. If user X visits products A, B, C and D and user Y visits products A, B, F and G, C and D are recommended to user Y and F and G are to user X.

The available algorithms are:

• Chef
• Complementary Products
• Highest Discounted Products
• Manual Merchandising
• Mixed Strategy
• Most Popular Items
• Most Valuable Products
• New Arrivals
• Purchased Together
• Purchased with Last Purchased
• Recently Viewed Products
• Substitute Products
• Top Sellers
• Trending Products
• User Based
• Viewed Together

Recommendation API

Recommendation API provides algorithm results through separated algorithm endpoints in JSON responses of which content has been stitched with your product data. 

Algorithms have required and optional parameters that can be used on recommendation strategies based on needs. Before implementation, make sure to have the desired algorithms activated by Insider team.