Predicting the future trend of popularity by network diffusion

An Zeng and Chi Ho Yeung

[Chaos 26, 063102 (2016);]

http://dx.doi.org/10.1063/1.4953013

在线网络中的趋势预测

从在线网络中预测出未来流行的商品在实际问题中有着重要的应用价值。传统方法主要基于商品已有流行性的线性外推方法来进行预测。这种方法在预测短期流行性方面有比较高的精确度，但在预测长期趋势时效果不佳。本文提出了一种基于用户-商品二分网上的扩散过程来做趋势预测。这种方法能够通过网络中微观层面的连边信息来预测商品的宏观行为。我们将这个方法运用在Netflix和Amazon这样的在线商务网络中，和美国物理学会引文网络中。结果显示我们的方法能够比线性外推法更准确的定位出小度商品中有潜力的商品（即未来将变得流行的商品）。

摘要：

Conventional approaches to predict the future popularity of products are mainly based on extrapolation of their current popularity, which overlooks the hidden microscopic information under the macroscopic trend. Here, we study diffusion processes on consumer-product and citation networks to exploit the hidden microscopic information and connect consumers to their potential purchase, publications to their potential citers to obtain a prediction for future item popularity. By using the data obtained from the largest online retailers including Netflix and Amazon as well as the American Physical Society citation networks, we found that our method outperforms the accurate short-term extrapolation and identifies the potentially popular items long before they become prominent.