Collaborative filtering (CF) is one of the most effective and popular recommendation methods. Deep learning has achieved great success in various tasks and extensive work has applied deep learning into collaborative filtering. However, existing deep learning based CF methods only capture either user-item relations (users' affection for target items) or item-item relations (association between target items and users' historical items), which may lead to sub-optimal recommendation accuracy. In this paper, we propose deep dual relations network, a neural network based approach that can model user-item relations and item-item relations simultaneously. Furthermore, considering the fact that a historical item a user prefers will have a more positive impact on his decision on a new item, user-item relations should play an important role in item-item relations modeling, which is not investigated by previous work. To solve this problem, we propose a special architecture, named affection-based attention network to capture the impact of user-item relations on item-item relations modeling. Comprehensive experimental results on four real-world datasets demonstrate the effectiveness of our proposed model.
By drawing an analogy between the population of an evolutionary algorithm and a gas system (which we call a particle system), we first build wave models of evolutionary algorithms based on aerodynamics theory. Then, we solve the models' linear and quasi-linear hyperbolic equations analytically, yielding wave solutions. These describe the propagation of the particle density wave, which is composed of leftward and rightward waves. We demonstrate the convergence of evolutionary algorithms by analyzing the mechanism underlying the leftward wave, and investigate population diversity by analyzing the rightward wave. To confirm these theoretical results, we conduct experiments that apply three typical evolutionary algorithms to common benchmark problems, showing that the experimental and theoretical results agree. These theoretical and experimental analyses also provide several new clues and ideas that may assist in the design and improvement of evolutionary algorithms.
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.