Question on Retrieval Stage assumption on Interactions

[harry7171]

Hi,

New to the TFRS, but going through basic guides and documentation . I have a question on the assumption of the Retrieval stage . for movielens dataset that all the data which is rated is treated as positive interactions. But for the use case of for example a different scenario where there is just a yes/no vote or like/dislike and the chances of positive responses is very sparse . how would the Retrieval work ?

For one of my usecases the positive interactions are very less .