Su Zhan, Zheng Xiliang, Ai Jun, Shang Lihui, Shen Yuming
School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, People's Republic of China.
Chaos. 2019 Aug;29(8):083133. doi: 10.1063/1.5099565.
The link prediction aims at predicting missing or future links in networks, which provides theoretical significance and extensive applications in the related field. However, the degree of confidence in the prediction results has not been fully discussed in related works. In this article, we propose a similarity confidence coefficient and a confidence measure for link prediction. The former is used to balance the reliability of similarity calculation results, which might be untrustworthy due to the information asymmetry in the calculation, and also makes it easier to achieve the optimal accuracy with a smaller number of neighbors. The latter is used to quantify our confidence in the prediction results of each prediction. The experimental results based on the Movie-Lens data set show that prediction accuracy is improved when the similarity between the nodes is corrected by the similarity confidence coefficient. Second, the experiments also confirm that the confidence degree of the link prediction results can be measured quantitatively. Our research indicates that the confidence level on each prediction is determined by the amount of data used in the corresponding calculation, which can be measured quantitatively.
链接预测旨在预测网络中缺失的或未来的链接，这在相关领域具有理论意义和广泛应用。然而，相关工作中尚未充分讨论预测结果的置信度。在本文中，我们提出了一种用于链接预测的相似性置信系数和一种置信度度量。前者用于平衡相似性计算结果的可靠性，由于计算中的信息不对称，该结果可能不可靠，并且还使得在邻居数量较少的情况下更容易实现最优精度。后者用于量化我们对每个预测的预测结果的置信度。基于Movie-Lens数据集的实验结果表明，当通过相似性置信系数校正节点之间的相似性时，预测精度会提高。其次，实验还证实了可以定量测量链接预测结果的置信度。我们的研究表明，每个预测的置信水平由相应计算中使用的数据量决定，而数据量是可以定量测量的。
