When building a hybrid recommender for products, how do you ensure that the model doesn't just recommender the most popular products all the time? Would throwing out X% (where X is high) of transactions involving this popular product be a good idea? With a hybrid model, could this most popular product just be left out of the data entirely?
1 Answer
By hybrid recommender, I assume you mean a system that weights both a user's profile/behavior and item content. By its very nature, a hybrid recommender will use item content which is independent of popularity. Collaborative filtering recommends popular items among users with similar profile/behavior, which is not the same as overall popularity.
-
$\begingroup$ Hmmm, this isn't an answer to my question. You correctly point out that the meta-data parts of the model will move the model away from just recommending top items, but what if my model still does anyway, i.e. because some items are just really popular? $\endgroup$ Aug 11, 2022 at 7:58