基于改进TADW的链路预测算法

针对经典的节点相似性链路预测算法只考虑网络拓扑结构或者节点属性信息的问题，使用词嵌入模型Word2vec学习得到节点文本属性信息的表示，进而改进TADW(text-associated deep walk)算法，弥补其语义信息表示能力的不足.基于改进的TADW图嵌入方法提出一种融合网络拓扑结构和节点属性信息的相似性指标，并基于此相似性指标提出链路预测算法.在三个真实数据集上的实验结果表明所提出算法可以提高预测精度，并具有更好的鲁棒性，同时使用图嵌入的方法有效解决了网络数据的稀疏性问题.

Link Prediction Algorithm Based on Improved TADW

Aiming at the problem that the classic node similarity link prediction algorithm only considers the network topology or node attribute information， the word embedding model Word2vec to learn the representation of node text attribute information is employed， and then TADW(text-associated deep walk)algorithm for its insufficient ability to express semantic information is improved. Based on the improved TADW graph embedding method， a similarity index which incorporate the topological structure and node attribute information is proposed. Furthermore， the link prediction algorithm is proposed based on this similarity index. Experimental results on three real datasets demonstrate the superiority of the proposed algorithm with better robustness on predicting precision as well as network sparsity solvability.