Improving Recommendation Services with Amazon SageMaker and MLOps: A Case Study with LotteON

In today’s digital age, personalized experiences are key to capturing and retaining customers. LotteON, a platform that offers a personalized shopping experience tailored to your preferred lifestyle, is setting the standard in this regard. They not only sell products but also provide personalized recommendations across various categories, including fashion, beauty, luxury, and kids.

To enhance the shopping experience even further, LotteON has been continuously improving its recommendation service. One of the ways they have done this is through the use of Amazon SageMaker and MLOps. By leveraging deep learning-based recommendation algorithms, such as Neural Collaborative Filtering (NCF), LotteON is able to analyze each customer’s unique tastes and needs.

The MLOps architecture that LotteON has built involves several components, including data preprocessing, automated model training and deployment, real-time inference through model serving, and a CI/CD structure. Each of these components plays a crucial role in ensuring that high-quality recommendations are provided to customers in real-time.

Data preprocessing is essential for handling the large amounts of data required for training the recommendation models. With the help of Amazon EMR, LotteON can process data quickly and efficiently. Automated model training and deployment streamline the process of training models and deploying them to production. Using SageMaker Pipelines, the team can define the steps required for ML services and manage the history of trained models and endpoints.

Real-time inference through model serving allows LotteON to provide recommendations to customers in real-time. By invoking the model deployed on a SageMaker endpoint, customers can receive personalized recommendations tailored to their preferences. Finally, the CI/CD structure ensures that updates to the recommendation models can be seamlessly integrated and deployed into production.

Overall, by utilizing Amazon SageMaker and MLOps, LotteON has been able to enhance the shopping experience for its customers. The deep learning-based recommendation models, such as NCF, have proven to be effective in providing high-quality recommendations. The MLOps platform that has been built enables quick development and experimentation with models, leading to enhanced recommendation experiences for customers.

If you’re interested in learning more about the NCF model and the MLOps configuration used by LotteON, you can check out their GitHub repo for hands-on practice. We hope that this post has given you insights into how to configure an MLOps environment and provide real-time services using AWS services. With the right tools and strategies in place, businesses can truly transform the customer experience and drive growth in today’s competitive market.

—
This blog post was co-written by SeungBum Shim, HyeKyung Yang, Jieun Lim, Jesam Kim, and Gonsoo Moon from LotteON. They are experts in data engineering, research engineering, and AWS solutions architecture, specializing in recommendation services and AI/ML technologies.