You are a machine learning specialist working for a social media software company. You have built and deployed a product recommendation model that recommends client products via social media posts in your company’s social media app. When you first deployed the model, it generated great results with users clicking through and buying client products, thereby generating revenue for your social media company. Over time the product recommendations results have started to decline, and your users are clicking through to client product pages less. You had not changed your model from when you did your initial deployment. What is the best and most efficient option to use to improve your user click-through rate for your social media app over time?

Answer options:

A.Periodically retrain your model using your foundational training data from your initial deployment adding new data from new client products.
B.Periodically retrain your model from scratch using your foundational training data from your initial deployment, adding an L1 or L2 regularization value set to the higher range of the parameter to represent client product changes.
C.Periodically update your model hyperparameters, setting the drift threshold to the higher range of the hyperparameter, to prevent model drift.
D.Completely recreate your model as it no longer recognizes client product changes.

Correct Answer: A
Option A is correct. Retraining your model with your initial training data plus new data representing new client products will keep the model in line with the product recommendation domain. This will allow your model to recognize new products while retaining the knowledge of the foundational product data.
Option B is incorrect. A regularization term will help prevent your model from overfitting, but it will not give your model the data it needs to recognize new product data.
Option C is incorrect. Model drift happens when you receive new data on which to train and your model. Changing your hyperparameters without retraining your model will not address your problem.
Option D is incorrect. Completely recreating your model is not necessary and definitely not the most efficient way to address your model performance problem. You will get better results more efficiently by adding new training data to your foundational training dataset and retraining your model.
References:
Please see the Amazon Machine Learning developer guide titled Training Parameters (https://docs.aws.amazon.com/machine-learning/latest/dg/training-parameters1.html),
The AWS News blog titled Amazon SageMaker Model Monitor – Fully Managed Automatic Monitoring For Your Machine Learning Models (https://aws.amazon.com/blogs/aws/amazon-sagemaker-model-monitor-fully-managed-automatic-monitoring-for-your-machine-learning-models/),
The Amazon SageMaker developer guide titled How Hyperparameter Tuning Works (https://docs.aws.amazon.com/sagemaker/latest/dg/automatic-model-tuning-how-it-works.html)