加权网络的在线结构学习算法

摘要

With continuous development of internet technology, the scope of network datasets increases massively. Analyzing the structure of network data is a research hotspot in machine learning and network applications. In this paper, a scalable online learning algorithm is proposed to speed up the inference procedure for the latent structure of weighted networks. Firstly, the exponential family distribution is utilized to represent the generative process of weighted networks. Then, using stochastic variational inference technique, the online-weighted stochastic block model ( ON-WSBM) is developed to efficiently approximate the posterior distribution of underlying block structure. In ON-WSBM an incremental approach based on the subsampling method is adopted to reduce the time complexity of optimization, and then the stochastic optimization method is employed by using natural gradient to simplify the calculation and further accelerate the learning procedure. Extensive experiments on four popular datasets demonstrate that ON-WSBM can efficiently capture the community structure of the complex weighted networks, and can achieve comparatively high prediction accuracy in a short time.%随着互联网技术的进步，网络关系数据不断涌现，规模不断膨胀，网络数据的结构分析成为机器学习和网络应用领域的研究热点。为了提高推理效率，文中提出加权网络的在线结构学习算法。首先，使用指数族分布描述加权网络的生成过程。然后，利用随机变分推理方法，构建加权网络的在线结构学习算法。该算法采用基于重采样技术的增量学习方式，降低优化的时间复杂度。最后，利用基于自然梯度理论的随机优化方法进一步加速学习过程，实现网络社区结构的在线学习和实时优化。通过与传统的离线学习算法进行对比实验，验证文中算法能高效快速地实现复杂加权网络的社区结构学习，并在较短时间内达到较高的预测精度。