during inference, is only CVR head used to rank items? or use CTCVR instead? or combine CVR head and CTR head(fuse both scores) together to rank items?
That really depends on your business use case. I'm not an expert in advertisement domain, but I know that there are several types of ads. Some ads require only high CTR and others require both CTR and CTCVR.
The goal of ESMM is to improve CTR and CVR prediction simultaneously. It will benefit all use cases.
thanks for this great post. I have a question:
during inference, is only CVR head used to rank items? or use CTCVR instead? or combine CVR head and CTR head(fuse both scores) together to rank items?
thanks!
That really depends on your business use case. I'm not an expert in advertisement domain, but I know that there are several types of ads. Some ads require only high CTR and others require both CTR and CTCVR.
The goal of ESMM is to improve CTR and CVR prediction simultaneously. It will benefit all use cases.
i see. I was just curious how to use this model in production as this model isn't necessarily tied to ads use case.