Knn

Created: 2022-03-23 11:35

Collaborative filtering method based on Similarity measures.
Simple and quite accurate, but suffers of low scalability and has problems in case of sparse databases -> not good for Cold-start.

User to user approach:

1. using the selected similarity measure, we produce a set of k nearest neighbors for the user;
2. aggregation phase: compute the average (or weighted sum) on item i;
3. take the best n items

Item to item approach:

1. produce the k items that are neighbors for each item in the database;
2. for each item i not rated by user a, compute its prediction using the ratings of a of the k neighbors of i;
3. select the top n recommenddations.

The two approaches (item-item and user-user) can be combined (paper)

References

1. https://www.sciencedirect.com/science/article/pii/S0950705113001044

Tags

#knn