Collaborative Filtering (Recommendation)

Definition

Collaborative Filtering (Recommendation)

Collaborative filtering is a recommendation approach that estimates a user’s preference for an unseen item from the explicit or implicit feedback of many users. Given users $U$, items $I$, and a partially observed user–item rating matrix $R$, it infers a missing entry

$${\hat{r}}_{ui}=F\left(R,u,i\right)$$

from similarities among users or among items. Its guiding assumption is that users who exhibited similar tastes in the past will tend to exhibit similar tastes in the future. It therefore uses the wisdom of the crowd rather than requiring a semantic description of each item. User-based nearest-neighbour collaborative filtering is one concrete form of this approach.