Spatially aware recommendations using k-d trees

Traditional Recommender Systems focus on recommending the most relevant items to users without considering any contextual features, such as time or location. In this work we propose a Recommendation Algorithm that takes user´s location into account while recommending. We focus on exploring the concept of spatial autocorrelation, i.e., similar values cluster together on a map, by using some statistical measures. This work uses a k-d tree based space partitioning technique to tessellate the users´ space with respect to location. Recommendations for the users are generated by combining their location and the preference statistics of other users that share the location. Our algorithm uses Collaborative Filtering, which is one of the widely used techniques for recommendation, by computing user-user or item-item similarities from the data. Since the Recommendation Algorithm is applied to each partition separately, we avoid the quadratic complexity typically associated with collaborative filtering. Our technique attempts to reduce the running time while ensuring that the quality of recommendations do not degrade. We have tested the algorithm on the MovieLens dataset. Experiments conducted indicate that our method is effective while reducing the running time.