attributes property

Extra user event features to include in the recommendation model.

If you provide custom attributes for ingested user events, also include them in the user events that you associate with prediction requests. Custom attribute formatting must be consistent between imported events and events provided with prediction requests. This lets the Retail API use those custom attributes when training models and serving predictions, which helps improve recommendation quality. This field needs to pass all below criteria, otherwise an INVALID_ARGUMENT error is returned: * The key must be a UTF-8 encoded string with a length limit of 5,000 characters. * For text attributes, at most 400 values are allowed. Empty values are not allowed. Each value must be a UTF-8 encoded string with a length limit of 256 characters. * For number attributes, at most 400 values are allowed. For product recommendations, an example of extra user information is traffic_channel, which is how a user arrives at the site. Users can arrive at the site by coming to the site directly, coming through Google search, or in other ways.