基于网络结构特征的相关用户推荐研究

On Related User Recommendation Based on Network Structure Analysis

推荐技术能够为用户提供符合其需求的资源,在信息过载的环境中推荐可有效增进传播效率。当前,传统推荐算法的效果在许多应用中得到证实,但在初始评价信息稀疏、数据量大的条件下该技术仍然存在不足。提出迭代资源扩散算法,通过用户在线行为构建用户评价网络,将此网络中节点结构相似性信息应用于相关用户推荐算法设计,可为网络用户提供与其具有较高相似性的相关用户列表;并进一步利用网络中节点的结构特征信息如节点度中心性、中介中心性等指标,改善算法效果。通过基于博客社区用户友情网络的对比测试实验,证明在初始信息非常稀疏的条件下,资源扩散算法推荐结果在覆盖度、命中率以及列表质量等指标上均优于传统推荐算法。

Recommendation systems can recommend related users who have similar online behavior before and provide proper resource, which is helpful in improving the efficiency of knowledge diffusion in an information overloaded environment. Thus, recommendation sys-tems can promote the practice of knowledge management. Some traditional recommendation algorithms have been proposed and tested by now, while there are disadvantages when the evaluation matrix is sparse and in large scale. Evaluation matrix can be built according to the users＆#39; online behavior, and the matrix can be transformed to a network. Based on structural similarity of actors in the network, the article proposes a recommendation algorithm, the iterated resource diffusion algorithm. Other structure indexes such as degree, betweenness cen-trality are also included to improve the algorithm. To evaluate the performance of the algorithm, the coverage rate, hitting rate and list quality are compared with traditional algorithms. The results show the advantage of the iterated resource diffusion algorithm.