• Project name: Web Fiction Recommender System

• Period: 2022-08 ~ 2022-12

• Role: Research, Develop and Serve Algorithm (model), Build training pipeline

• Key skills: BERT, GRU, SageMaker with custom container, Airflow ETL

• Result: Content Consumption - improved 4.30%+, Retention Rate - improved 4%p+, Average Precision@25 40%

• Sample code: https://github.com/qordmlwls/sequential_recommender_system_sagemaker_with_custom_container

• Supplementary:

  Recommender system with SageMaker_ver 2.0.pptx

[Contents]

Motivation

• Overcome limitations of a former recommender system based on Amazon Personalize
• Construct a recommender system using content embedding data and users’ sequential interest from their historical behaviors
• Deploy recommendation model using Amazon SageMaker

MLOps process

• MLOps process for reference