Amazon Recommendations: Ranking Items & Matching Buyers
Reference article:
http://www.martechadvisor.com/articles/customer-experience/recommendation-engines-how-amazon-and-netflix-are-winning-the-personalization-battle/
Personalization of recommendations is extremely important in the e-commerce industry. According to the article linked above, recommendations are responsible for roughly 35% of sales. The goal is to show the relevant items and at the right times, and expose the customers to products they otherwise would not have discovered.
In class we discussed Google page ranks and how this idea improves the user experience on the search site. Analogously, Amazon ranks its multitude of products so that it can accurately match them to potential customers. In short, Amazon gathers data from 3 main elements:
- purchase history
- items in the shopping cart
- viewed items and what other customers viewed and purchased
Then, this data is used to rank the potential items in the right order so that it is as relevant as possible for each individual customer.
Amazon uses “item-to-item collaborative filtering” and recommendation engines that learn user preferences. The customer experience is personalized using data science and machine learning. Furthermore, deep learning, which involves a large network of computing power to carry out even more complex forms of machine learning increases the efficiency of the overall process. In fact, Amazon uses an AI framework called DSSTNE that is open source.
Most web-based companies really invest in this area to keep improving their results so that they stay competitive. They strive to improve their AI technology and ultimately get better recommendation results while decreasing the amount of data needed. All this effort is to increase the chances that a customer will click on an item and buy it.