Recommendations of Unique Items Based on Bipartite Graphs
Kristóf Marussy, Ladislav Peška, Krisztián Búza · Repository of the Academy's Library (Library of the Hungarian Academy of Sciences) · 2015
We consider a model of recommending items to users based on weighted bipartite graphs for domains where the availability of items is limited, i.e.only a fixed number of users can acquire a single item.As a motivating example, we consider the problem of an electronic antique book store where each book can be sold to only a single customer.Three approaches are used for recommendation generation, including greedy top-k recommendation and its generalizations k, ℓ-recommendation and k, W -recommendation.We show that the k, W -recommendation problem is NP-complete.The recommendation methods are subjected to data-driven and stochastic evaluation.A stochastic model is used to quantify user dissatisfaction due to desired items going out of stock.Our numerical experiments have shown that greedy recommendation is outperformed by k, ℓ-recommendation and k, W -recommendation in the antique book store recommendation problem both in terms of accuracy and (expected) user satisfaction.The more principled selection of parameters for the recommendation algorithms and the stochastic model is scope of future work, as well as applications to other domains.