Obtaining feature importance/Sensitivity for model interpretability

Nobody has claimed this yet.

Hi, I'm new to recommender models and LightFM . I'm creating model for customer like/dislike recommendations (no ratings involved). What are the available options for model interpretability in such cases?

I understand LightFM is a hybrid approach, but is there a way I can rank user/item features on the basis of importance in the model predictions. Or understand the impact of user/item features on predictions. In regular ML models I can assess this using permutation feature importance, partial dependency plots for example.

Please let me know if someone has done a related analysis or if there's just no way for model interpretability in this case.

The issue names no files, tests, or entry points. Start by reviewing LightFM's model and prediction APIs to determine whether feature importance or sensitivity can be defined, then document the proposed scope and how any interpretation output would be validated.

Written by the indexing model from the issue text.