Hybrid Recommenders
Best for: Combining user, item, and context signals
How it works
$$\hat{r}_{ui}=\alpha\,\hat{r}^{CF}_{ui}+(1-\alpha)\,\hat{r}^{CB}_{ui}$$Combines collaborative and content-based signals so content features cover cold-start items and collaborative signals provide serendipity. The simplest form is a weighted blend $\hat{r}_{ui}=\alpha\hat{r}^{CF}_{ui}+(1-\alpha)\hat{r}^{CB}_{ui}$, but production systems more often feed user, item, and context features into a single model such as a factorisation machine, gradient-boosted trees, or a deep network so that all evidence informs one score. A common architecture stacks a ranker on top of several candidate generators (collaborative, content, graph).
Common fields
Most production recommender systems