We employ a bipartite network to describe an online commercial system. Instead of investigating accuracy and diversity in each recommendation, we focus on studying the influence of recommendation on the evolution of the online bipartite network. The analysis is based on two benchmark datasets and several well-known recommendation algorithms. The structure properties investigated include item degree heterogeneity, clustering coefficient and degree correlation. This work highlights the importance of studying the effects and performance of recommendation in long-term evolution.The digital revolution has brought to us the socalled "information overload": there is too much information for online users to deal with. As a result, nowadays there is hardly an e-commerce website without some form of information filtering or recommendation service. Recommender systems seek to predict users' non-considered preference typically through collaborative or content-based filtering.[1−4] Researchers have developed lots of methods to improve the effectiveness of recommendation, such as matrix factorization, [5] restricted Boltzmann machines, [6] social tags [7] and the ensemble method.[8]Online commercial systems can be well described by bipartite networks [9] where users and items are represented by nodes, and an edge means that a user has selected an item. In recent years, the network-based recommendation methods became the focus in the literature. For example, the mass diffusion [10] and heat conduction [11] are two personalized recommendation algorithms based on the diffusion process on bipartite networks. A hybrid algorithm of them is shown to effectively solve the accuracy-diversity dilemma.[12]The network manipulation method is also introduced to improve the recommendation performance.[13]Even studied intensively, most previous works concentrate on evaluating the performance of single recommendation based on training-probe set division. It is not sure that a well-performed recommendation method in a single step can enjoy high performance in the long term. Some methods may make the online network evolve to an unhealthy state. For example, the whole market might be dominated by several super popular items thus users only have limited choices. On the other hand, items may have very even popularity, which makes the quality of the objectives indistinguishable from the degree. Therefore, it is important to study how different recommendation algorithms affect the structure of an online network in long-term evolution.Fortunately, the tremendous wave of research on complex networks [14−17] in the past decade provides us with a powerful tool to uncover the structure properties and function of the bipartite network. Due to the wide existence of the bipartite network in reality, such as human sexual network [18] and collaboration network, [19] great effort has been made to study its empirical analysis, [20] node strength connectivity correlation, [21] projection into monopartite network [22] and topology metrics. [23] In this Letter, w...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.