question

50919478 avatar image
0 Votes"
50919478 asked GiftA-MSFT commented

"Cold-start" problem in Azure Recommender

Hi, I've trained and deployed an Azure Wide & Deep Recommender in the Designer tab of the Machine learning workspace.
I have 3 datasets:

  • Ratings

  • User features

  • Item features

The recommendation system works fine for existing user IDs available in User Features and Rating datasets.
However, when a want to make a prediction for a new user to the system, which IDs were not used during training, I get an error.

Is that expected that the recommendations are made only for users that the model learned during training? If not, please help me resolve this issue, so that even "cold" users can receive recommendations.

The model training and real-time inference pipelines are attached. Also, included the deployment logs with errors when sending a request to the model to make recommendations for new users.

103326-ci-error-logs-recommender-system.txt
103308-model-training-pipeline.png
103320-model-training-pipeline.png



Thanks!

azure-machine-learningazure-machine-learning-inference
· 1
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

Hi, you should be able to create scores for new users using collaborating filtering data. Can you share your process and configuration for scoring new data? Did you ensure to provide a dataset containing user-item-ratings as well as user and item features during prediction?


0 Votes 0 ·

0 Answers