基于LDA主题模型的社会网络链接预测

针对传统社会网络链接预测方法忽视节点文本内容的问题,提出一种基于潜在狄利克雷分配(Latent Dirichlet Allocation, LDA)主题模型的协作演化链接预测算法。算法利用LDA模型,对节点的文本内容进行分析,提取出每个节点的主题分布向量,利用分布向量的点积来衡量节点文本的相似性;然后将节点文本内容相似性矩阵与节点邻接矩阵相加,在此基础上计算节点之间的相似性;最后选取相似性最高的k个节点作为预测结果。实验结果表明该算法在网络图稀疏的情况下有较好的效果。

LDA-based link prediction in social network

To address the problem of ignoring the text contents of nodes in social network link prediction methods, a Latent Dirichlet Allocation(LDA)-based collaborative evolutionary link prediction algorithm was proposed. The algorithm used LDA model to analyze the text content and abstracted a topic distribution vector for each node; The product of the topic distribution vectors was adopted to measure the similarity between the nodes' contents; Afterwards, the content similarity matrix was added to the adjacency matrix and the similarities between the nodes were computed consequently; At last, k most similar nodes were selected as the prediction result. The experimental results showed that the proposed algorithm achieved good prediction performance in sparse networks.