文献类型：期刊文献

中文题名：二分网络中多步物质扩散推荐算法的逼近分析

英文题名：Approximation Analysis of Multi-Step Material Diffusion Recommendation Algorithm Based on Bipartite Network

作者：周海平[1];沈士根[1];黄龙军[1]

机构：[1]绍兴文理学院计算机科学与工程系

年份：2018

卷号：47

期号：3

起止页码：436

中文期刊名：电子科技大学学报

外文期刊名：Journal of University of Electronic Science and Technology of China

收录：CSTPCD、、EI(收录号：20183505757409)、北大核心2017、Scopus、CSCD2017_2018、北大核心、CSCD

基金：国家自然科学基金(61772018);贵州省科学技术基金(LH[2014]7210;LH[2015]7294)

语种：中文

中文关键词：逼近分析;二分网络;扩散转移矩阵;物质扩散;推荐算法

外文关键词：approximation analysis;bipartite network;diffusion transfer matrix;material diffusion;recommendation algorithm

中文摘要：物质扩散推荐算法由于其简洁有效的特点自提出以来便受到了广泛关注。然而,至今为止,人们对该算法的研究大多停留在二分网络中的两步扩散过程,而对多步扩散过程少有涉及。该文利用矩阵分析的方法对二分网络中的多步物质扩散过程进行了研究,通过对扩散转移矩阵W的性质进行分析,发现当扩散步数N趋于无穷时WN收敛。与之对应,在多步扩散之后,网络中的资源分布最终会达到一个稳定状态,且此时每个节点最终获得的资源数与该节点的度成正比,而与资源的初始分布状态无关,这导致推荐算法将完全根据物品的热门程度进行推荐,推荐结果也不再具有个性化特点。研究结果表明二分网络中的物质扩散推荐算法随着扩散步数的增加将逐渐失去个性化特性。

外文摘要：Material diffusion recommendation algorithm has received wide attention for its simplicity and effectiveness. However, up to now, most of researches on this algorithm were confined to two-step diffusion process of a bipartite network. In this paper, we use the method of matrix analysis to study the multi-step material diffusion process in the bipartite network. By analyzing the nature of the diffusion transfer matrix of W, we prove that WN converges when the diffusion step N tends to infinity. Eventually, the distribution of resources will reach a steady state. At this point, the number of resources that each node obtains is proportional to its degree, but not the proportion of the initial distribution of resources. At the same time, the algorithm is transformed into a global recommendation algorithm and the recommendation result is no longer personalized. It reveals that the material diffusion recommendation algorithm will gradually lose its personalized feature with the increase of the number of diffusion steps.