算法在链路预测与错边识别中鲁棒性差异分析
n analysis for the difference of algorithms\' robustness between link prediction and spurious link elimination
随着信息采集和处理的技术不断发展，现实生活中我们采集到的数据规模呈现爆发式的增长，然而我们采集到的实际数据中往往存在着噪声数据。复杂网络中的链路预测和错边识别问题均是数据挖掘领域重点研究的方向，在众多领域的真实系统中两者都有实际应用的价值和案例，因此众多的算法被不断提出。然而在解决实际问题时，算法的应用场景主要都是在充满着噪声的环境中，那么算法在有噪环境下预测或识别错误边的准确度会受到怎样的影响，算法的鲁棒性问题是本文关注和考察的重点。错边识别与链路预测问题类似但又有所差别，人们往往不加区分地将针对于链路预测所提出来的算法直接应用于错边识别中，那么算法在链路预测和错边识别问题中所呈现的鲁棒性差异是怎样的呢。针对这一问题，本篇文章考虑了网络中存在噪声连边的情况，首先分析了错边识别中算法的鲁棒性，然后对比了算法在链路预测和错边识别中呈现的鲁棒性差异。本篇文章通过探究随噪声比例的改变，算法AUC值的变化趋势间接的体现算法鲁棒性变化情况。最后我们发现，算法在链路预测问题中对噪声较敏感，而在错边识别中表现出较强的鲁棒性。
Nowadays a lot of data sets have been accumulated in various and wide fields due to the information collection and processing technology. However, the actual data we obtain often contain noise links.The link prediction and identifying missing interactions in complex network both havean increasingly wide utilization in various fields. Therefore, many algorithms have been put forward.However,in practice, the application scenarios of algorithms are mainly in the noisy environment.In this paper, we computes the robustness of the algorithms in noisy environment to study the influence of the algorithm is affected by the noise. In practice, people directly apply the link prediction algorithms to identify spurious links despite the performance difference between two different applications. In this paper, we computes the robustness of the algorithms in recognition spurious links and compares it with the link prediction.By investigating the trend of the AUC values with the varing of the ratio of noise. Finally,we find that the 19 algorithms are more sensitive to noise in link prediction, and shows stronger robustness in the recognition spurious links.
肖井华、张鹏、邱丹
计算技术、计算机技术
复杂网络链路预测错边识别鲁棒性
omplex networkLink predictionIndentifyingSpurious linkRobustness
肖井华,张鹏,邱丹.算法在链路预测与错边识别中鲁棒性差异分析[EB/OL].(2018-01-04)[2025-08-04].http://www.paper.edu.cn/releasepaper/content/201801-19.点此复制
评论