Distributed Implementation of Latent Rating Pattern Sharing Based Cross-domain Recommender System Approach

Latent rating pattern sharing based approaches for cross-domain recommendations can alleviate the data sparsity problem by pulling the knowledge available from other domains and are faster in prediction. However, since the prediction quality depends on number of chosen user and item classes for given data-set, the model training time becomes prohibitively large even for medium size data-sets. In this paper, we propose a MapReduce based distributed implementation of the cross domain recommendation algorithm. Our implementation has the capability to run on modern distributed computing frameworks, such as Hadoop and Twister, that utilize commodity machines. The experimental results show that the training time increases only linearly with user and item classes when compared to the exponential increase in case of its sequential counterpart.