Product Recommender for E-Commerce
E-commerce website swamps you with recommendations on what to buy. Have you ever wondered why you see what you see as recommendations? These recommendations are generated by product recommenders that uses predictive analytics at core. At very high-level, there are four types of recommenders.
Types of Recommenders
Most Popular Items: The first type of recommender is ‘the most popular items’ filter. This is the simplest recommender and does not require machine-learning algorithms. It simply looks at sales data to recommend best sellers or viewing history to recommend most popular items.
Customer Preferences: The second type of recommender is the ‘content-based filters (CBF)’. CBF tracks a user’s actions, such as order history, browsing history, time spent in various product categories, etc. It uses this data to create a customer profile and matches the profile with product attributes. CBF recommends product with highest matching attributes with customer’s preferences and behavior. This recommender helps to generate real-time personalized recommendations based on shopper’s current and past viewing and purchasing behavior. This recommender helps to implement product recommendation strategies, such as ‘Related to items you have viewed’, ‘recommended for you’ ,‘recently viewed’, ‘buy it again’ and ‘inspired by your purchase trends’.
Semantic preferences: The third type of recommender is the ‘collaborative filters (CF)’. CF looks at user’s behavior and preferences. It predicts what each user would like/purchase based on the user’s similarity to other users. This recommender helps to establish recommendation strategies, such as ‘customers who bought also bought’, ‘customers who viewed also viewed’ and ‘customers who viewed ultimately bought’.
Purchase basket: Lastly, there is a type of recommender called the complementary filtering that identifies most frequent co-purchase. The system learns the probability of two or more products being bought together. It generates product recommendations based on product affinity. The recommendation strategies supported by complimentary filtering are ‘frequently bought together’ and ‘products related to this item’.
References:
1. <https://www.datasciencecentral.com/profiles/blogs/understanding-and-selecting-recommenders-1>
2. <https://www.netguru.com/blog/product-recommendation-machine-learning>
3. <https://www.datasciencecentral.com/m/blogpost%3Fid%3D6448529%3ABlogPost%3A512183>
4. <https://developers.google.com/machine-learning/recommendation/collaborative/basics>
5. <https://www.amazon.science/blog/improving-complementary-product-recommendations>
6. <https://towardsdatascience.com/association-rules-2-aa9a77241654>
7. <https://towardsdatascience.com/complete-guide-to-association-rules-2-2-c92072b56c84>
9. Davis, Melissa. The future of customer service analytics. ID: G00403423 <https://www.gartner.com/document/3964483?ref=solrAll&refval=267449578>