Research on Influence Maximization Algorithm and Recommendation Application based on Social Big Data

在线社交网络中的个体与复杂的关系联系在一起，这导致了社交网络的复杂特征。用户影响力在信息传播过程中发挥着重要作用。影响力指的是一种改变他人行为和思想的重要能力。我们一直致力于推出一些新的分析方法来研究社交大数据中的用户影响力。我们在2020年开展了如下研究：首先，利用流体动力学建立了一个新的影响力扩散模型，揭示了影响力扩散的时间演化过程。建立了正面影响力最大化问题的数学模型，设计了一种贪婪算法--流体扩散算法进行求解，并建立了基于信任的竞争影响力扩散模型来模拟正面影响力和负面影响力的传播。通过启发式剪枝方法，提出了一种高效的基于信任的竞争影响力最大化算法，并在此基础上提出了一种端到端的学习影响参数改进方法--多维影响向量法。它学习双任务网络嵌入，以联合预测影响概率和级联大小。

Individuals in online social networks are linked with complicated relationships that lead to the complex characters of social networks. Users’ influence plays an important role in the process of information diffusion. The influence denotes an important ability that changes the behavior and thoughts of other people. We have been focusing on deriving some new analysis methods to study user influence in social big data. We carried out the study in 2020 as follows:First, we established a new model of influence spread using fluid dynamics, which reveals the time-evolving process for influence spread. The problem of maximizing positive influence was formulated and a greedy algorithm, Fluidspread, was devised to solve the problem.Then, a model of trust-based competitive influence diffusion was established to simulate the spread of positive and negative influence. An efficient algorithm of trust-based competitive influence maximization was developed through a heuristic pruning method.Finally, an end-to-end method was devised that uses dual-task network embeddings to improve learning influence parameters, which is called a multi-dimensional influence-to-vector method. It learns dual-task network embeddings to jointly predict influence probabilities and cascade sizes.