INSUFFICIENT_TRAINING_DATA while meeting data requirements

[eetsenl]

Hi,

It seems the model cannot be trained because of insufficient data. However in the console interface it says all requirements are met. See screenshot below.

I have 1000+ documents with the default media schema.

Any suggestions on how to fix this? Any help would be appreciated.

Kind regards,

Ben

[MarvinLlamas]

Hi @eetsenl,

Welcome to Google Cloud Community!

The message 'Data requirements met' is misleading. While your view-item data is sufficient, the media-play data is not. It looks like you need sufficient media-play data to meet either the view-item or media-play requirements, along with the documents requirement, to train the model. The view-item data is already provided and won't add any value.

To address your question, here are potential ways that might help with your use case:

• Review your tracking: You might want to review your media-play event implementation to ensure the events trigger properly during user interactions with the media.
• Conduct Thorough Testing: Once tracking is set up, you may thoroughly test media interactions to verify that the data is being captured accurately.
• Allow Time for Data to Accumulate: After implementation, you may wait for the data to collect and appear on the dashboard.
• Check your Dashboard: After collecting sufficient media-play event data and resolving the issue,  all requirements should be fulfilled, and you'll be ready to train the model.

You may refer to the following documentation, which explains the data sufficiency issues, clarifies why the model can't be trained with just view-item data, and provides guidance on correctly implementing media-play events: 

• About user events

Was this helpful? If so, please accept this answer as “Solution”. If you need additional assistance, reply here within 2 business days and I’ll be happy to help.

[eetsenl]

Hi @MarvinLlamas,

Thanks for your thorough reply. Really appreciate it!

I was under the impression that you need either view-item events or media-play events. Not both. This is also what the data requirements page seems to imply ("OR").

We have a recipe website and want to recommend recipes on the recipe detail page. However most of the recipes don't have any video content. Is this the wrong use case for media recommendations? Or could it be solved, for example, by always firing a media-play and media-complete event.

For now we have switched to generic recommendations and this is working ok. In the docs it says that user events are not required for this. However, it's not really clear whether or not user events are actually used to train the model and personalize recommendation results. When I check the results as a different user they always seem to be the same.

Looking forward to your reply.

Kind regards,

Ben