DeepRank: Learning to rank with neural networks for recommendation
DeepRank: Learning to rank with neural networks for recommendation
复制标题
DOI：
10.1016/j.knosys.2020.106478
复制
发表时间：
2020-12-17
影响因子：
8.8
通讯作者：
Zhou, Xiuze
中科院分区：
文献类型：
--
作者：
Chen, Ming;Zhou, Xiuze
虽然广泛应用的深度学习模型在推荐系统中表现出良好的性能，但很少有人致力于探索推荐系统中的排名学习。推荐系统的最终目标是为用户推荐一个经过排序的项目列表，生成一个高质量的排序列表对于推荐系统来说非常重要。此外，从基于矩阵分解（MF）的方法学习的潜在特征没有考虑潜在特征之间的任何深层交互;因此，它们不足以捕获用户项潜在结构。为了解决这些问题，我们提出了一种新的模型，DeepRank，它使用神经网络来提高协同过滤（CF）的个性化排名质量。这是一个通用的架构，不仅可以很容易地扩展到进一步的研究和应用，但也可以简化成对学习排名。最后，我们在三个数据集上进行了广泛的实验。结果表明，我们提出的模型显着优于国家的最先进的方法。我们的项目可在：https://github.com/XiuzeZhou/deeprank. (C)2020爱思唯尔B. V.保留所有权利。
Although, widely applied deep learning models show promising performance in recommender systems, little effort has been devoted to exploring ranking learning in recommender systems. It is important to generate a high quality ranking list for recommender systems, whose ultimate goal is to recommend a ranked list of items for users. Also, the latent features learned from Matrix Factorization (MF) based methods do not take into consideration any deep interactions between the latent features; therefore, they are insufficient to capture user-item latent structures. To address these problems, we propose a novel model, DeepRank, which uses neural networks to improve personalized ranking quality for Collaborative Filtering (CF). This is a general architecture that can not only be easily extended to further research and applications, but also be simplified for pair-wise learning to rank. Finally, we perform extensive experiments on three data sets. Results demonstrate that our proposed models significantly outperform the state-of-the-art approaches. Our projects are available at: https://github.com/XiuzeZhou/deeprank. (C) 2020 Elsevier B.V. All rights reserved.