Abstract: With the continuous emergence of various social networking sites, finding a group of the most influential users on multiple social networks is very important for product recommendation or product promotion. In order to improve the breadth and accuracy of product recommendation or promotion, this paper presents an algorithm of topic-aware influence maximization, M-TLTGreedy. Firstly, this paper evaluates the relation among users based on their text semantics and social relationships in multiple social networks to build a topic-based cross-network graph. Then, based on the linear threshold model, this paper designs a topic-aware influence maximization model across social networks, M-TLT (multiple-topic linear threshold) model. Next, this paper uses the improved heuristic algorithm to select a set of users based on the M-TLT model. Finally, the extensive experiments on real datasets show that the   M-TLTGreedy algorithm performs well on influence spread and running time.

Key words: social networks, topic-aware, influence maximization, linear threshold model

摘要： 随着各种社交网站的不断涌现，在多社交网络上找到影响传播范围最大的一组用户，对产品推荐或产品推广具有重要作用。为提高产品推荐或推广的广度和精准性，提出了一种跨社交网络基于话题感知的影响力最大化处理方法M-TLTGreedy。首先，根据跨社交网络中的文本语义信息和用户间的社会关系来评价多社交网络中用户间关系，以此构建一个基于话题的跨社交网络图；然后，在线性阈值模型的基础上，设计了一个基于话题感知的跨社交网络影响力最大化模型M-TLT（multiple-topic linear threshold）；接着，基于M-TLT模型，利用改进的启发式算法，进行初始用户集的选取；最后，基于大量数据集的实验，证明了该算法无论在影响范围和时间效率上均表现良好。

关键词: 社交网络, 话题感知, 影响力最大化, 线性阈值模型