In this repository, we will build a real-time recommendation engine for an e-commerce website using a synthetic online grocer dataset. We will use Amazon SageMaker Feature Store(both online and offline) to store feature data that will be used for model training, validation and for real-time inference. The recommendation engine will suggest top products that a customer is likely to purchase while browsing through the e-commerce website based on the customer's purchase history, real-time click stream data, and other customer profile information. The solution is a combination of two ML models - a Collaborative Filtering model and a Ranking model.
We will use a collaborative filtering model based on matrix factorization using the Factorization Machines Algorithm to retrieve product recommendations for a customer. This will be based on customer profile and past purchase history of a customer in addition to features such as product category, name, and description. The customer's historical purchase data and product data from the e-commerce store's product catalog are stored in two separate offline Feature Store feature groups that will be combined to train this model.
We will also train an XGBoost model based on click stream historical aggregates data to predict a customer's propensity to buy a given product. We will use aggregated features on real-time click stream data (stored and retrieved in real-time from the SageMaker Feature Store) along with product category features. We will use Amazon Kinesis Data Streams to stream real-time click stream data and Amazon Kinesis Data Analytics to aggregate the streaming data using a stagger window query over a period of the last 2 minutes. This aggregated data will be stored into an online Feature Store Feature Group in real-time to be subsequently used for inference by the ranking model.
Original authors:
- Bobby Lindsey
- Vikram Elango
- Anjan Biswas
- Arnab Sinha
See CONTRIBUTING for more information.
This library is licensed under the MIT-0 License. See the LICENSE file.