Bloofi Representation for Item/User in Recommender Systems

摘要

In this paper, we propose new algorithms to improve the performance of recommender systems, based on hierarchical Bloom filters. Since Bloom filters can make a tradeoff between space and time, proposing a new hierarchical Bloom filter causes a remarkable reduction in space and time complexity of recommender systems. Space reduction is due to hashing items in a Bloom filter to manage the sparsity of input vectors. Time reduction is due to the structure of hierarchical Bloom filter. To increase the accuracy of the recommender systems we use Probabilistic version of hierarchical Bloom filter. The structure of hierarchical Bloom filter is B+ tree of order d. Proposed algorithms not only decrease the time complexity but also have no significant effect on accuracy.