采用群体信息的二部图链接预测方法

Bipartite Graph Link Prediction Method Using Community Information

二部图包含2种不同类型的节点且链接只存在于不同类型的节点之间,因此,许多适用于普通单部图的链接预测方法无法直接用于二部图中。另外,群体信息对提高链接预测的准确率有重要意义,但缺乏相关研究。为此,提出一种采用群体信息的二部图链接预测方法。将链接预测视为机器学习的分类问题,通过对二部图投影,抽取二部图中节点对样本的局部结构属性,并运用群体检测技术抽取节点对样本的群体属性,并把局部结构属性和群体属性一起作为节点对相似度的度量标准,在监督学习框架中进行训练和预测。在现实数据集Movie Lens中的实验结果表明,群体信息的引入能有效提高二部图链接预测方法的准确率,改善推荐性能。

Since bipartite graph contains two different types of nodes and links only exist between different types of nodes, most link prediction methods for common single graph cannot be applied to bipartite graphs directly. In addition, the community information does not draw enough attention though it is important to improve the accuracy of link prediction. So a bipartite link prediction method using community information is developed. The method regards link prediction as a classification problem in machine learning. The local structural properties of node pair instances in a bipartite graph are extracted by the projection of the bipartite graph, together with community properties of instances by exploiting community detection techniques. Then local structural properties and community properties are used as similarity measurements of node pair instances. Training and prediction are conducted in a supervised learning process. Experimental results in a real dataset MovieLens show that the use of community information improves the link prediction accuracy and recommendation performance.