Seeking advice on retrieval-augmented classification for seasonal prediction tasks

Follow the full discussion on Reddit.
I'm working on a project to train a binary multi-modal classifier for predicting political content. Since political content tends to have seasonal trends, I want to use a retrieval-augmented classification setting. This way, whenever a new trend emerges, I can incorporate new features into my retrieval dataset and improve the model's precision. Additionally, I'd like the ability to override the model's decisions based on high similarity in the retrieval dataset. Can anyone recommend relevant papers or techniques for this approach? Any guidance or resources would be greatly appreciated!