Finding the Positive Nearest-Neighbor in Recommender Systems

Recommender systems make suggestions about products or services based on matching known or estimated preferences of users with properties of products or services (contentbased), or properties of other users considered to be similar (collaborative filtering). Collaborative filtering is widely used in Ecommerce. To generate accurate recommendations, the properties of a new user must be matched with those of existing users as accurately as possible. The available data is very large, and the matching must be computed in real time. We introduce algorithms that use “positive” nearest-neighbor matching in the sparse datasets typical of collaborative filtering to find near neighbors whose attribute values exceed those of each new user. The algorithms use singular value decomposition as a dimension-reduction technique. Making this idea effective requires careful attention to details such as normalization. Experimental results are reported for a movie recommendation dataset. For n users and m objects, reasonable matches can be found in time O(m log n), using O(nm) storage.