选择PageRank阻尼因子的奇异扰动法
A singular perturbation approach for choosing PageRank damping factor
摘要：
PageRank阻尼因子的选择并不明显。Google选择c=0.85的值是在真实反映Web结构和数值效率之间的折衷。然而，原始Web图上的马尔可夫随机游动并不能反映页面的重要性，因为它吸收了死胡同。因此，阻尼因子不仅需要加速计算，而且还需要建立页面的公平排名。在本文中，我们提出了新的标准，选择阻尼因子的遍历结构的Web图和概率流的基础上。具体来说，我们要求核心组件获得PageRank质量的公平份额。利用奇异扰动方法，我们得出结论，值c=0.85太高，并建议阻尼因子应选择在1/2左右。作为副产品，我们详细描述了Web Graph的OUT组件的遍历结构。我们的分析结果证实了两个大样本的Web图的实验。
英文摘要：
The choice of the PageRank damping factor is not evident. The Google's choice for the value c=0.85 was a compromise between the true reflection of the Web structure and numerical efficiency. However, the Markov random walk on the original Web Graph does not reflect the importance of the pages because it absorbs in dead ends. Thus, the damping factor is needed not only for speeding up the computations but also for establishing a fair ranking of pages. In this paper, we propose new criteria for choosing the damping factor, based on the ergodic structure of the Web Graph and probability flows. Specifically, we require that the core component receives a fair share of the PageRank mass. Using singular perturbation approach we conclude that the value c=0.85 is too high and suggest that the damping factor should be chosen around 1/2. As a by-product, we describe the ergodic structure of the OUT component of the Web Graph in detail. Our analytical results are confirmed by experiments on two large samples of the Web Graph.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
