Research On Role-based Representation Learning And Influence Maximization In Multiplex Networks

Social media has been integrated into people’s daily life and work,and plays an important role in society,and the economy.Its huge user base has given birth to many businesses based on social networks,such as viral marketing,and word-of-mouth marketing.These applications rely on the spread of information in social networks,so how to find the most influential spreader and promote the widespread diffusion of information becomes crucial.This is the core of the research on Influence Maximization(IM),in which the initial spreaders of information are called seed users.The IM problem in single-layer networks has been extensively studied,but these studies ignore the multiplicity of social networks.The traditional social network adopts a binary modeling method,which only considers whether there is an interactive relationship between users,but ignores the diversity and importance of the interactive relationship.Therefore,the multiplex network is more suitable for describing these complex real situations.Therefore,the traditional IM problem becomes the IM problem in multiplex networks,which also makes the problem more complicated.In recent years,the seeding problem in single-layer networks has been tackled using network representation learning methods,owing to their proficient feature extraction capabilities.But for multiplex networks,there are few studies in this area.In addition,because information does not exist in isolation in the network,its diffusion process is often affected by competitive and cooperative information.Therefore,it is also necessary to consider the influence of different information in the process of influence spreading.Given the above situation,this paper aims at the IM problem in multiplex networks,from the perspective of seed node selection and influence spread model,and conducts research from multiplex network representation learning methods and multiple information coupling diffusion models.Then,based on these two aspects of research,IM problems in multiplex networks are discussed.Primarily,the seeding problem is approached through the utilization of multiplex network embedding methodologies.Additionally,an influence propagation model is devised by incorporating competitive and cooperative information effects,building upon the foundation of research focused on multiple information-coupling propagation.The research work of this paper mainly includes the following aspects.(1)A network representation learning method based on role-aware random walk is proposed.The proposed random walk method can flexibly capture the community information and structural role information of nodes to adapt to different downstream tasks.For single-layer networks,sampling at the role level enables the proposed method to be applicable to different network structures,including disconnected networks.For multiple networks,same-layer and cross-layer context sampling can be achieved through role-aware random walks.On this basis,the integration of the Skip-gram model enables the learning of node representations for both single-layer and multiplex networks.Empirical evaluations on multiple real-world datasets substantiate the efficacy of the proposed method.(2)A unified multiplex network representation learning framework is proposed,which incorporates different structural information in multiplex networks.For representation learning of multiplex networks,the proposed learning framework is able to unify structural role information into one learning framework.In addition to preserving node proximity,the learned node representation can also preserve the structural similarity between nodes.Further,in a semi-supervised manner,the importance of node representations at different layers is learned through an attention mechanism.Comparisons with baseline methods on multiple public datasets demonstrate the effectiveness of the proposed method.At the same time,the ablation experiments also demonstrate the enhancement effect of structural role information on multiplex network embeddings.(3)A coupled spread model of the existence of competitive and cooperative information in multiplex networks is constructed.The proposed framework unifies competitive and cooperative coupled spread processes,taking into account the impact of both competitive and cooperative information in the model.The theoretical threshold of information is analyzed through the Microscopic Markov Chain Approach(MMCA),and a large body of numerical simulations is used to discuss the information coupling spread process in multiplex networks.The research results reveal the inhibitory and promoting effects of competitive information and cooperative information on target information respectively,and also reveal the indirect effect between information.This study deepens the understanding of the coupled information spread process in multiplex networks,and provides a theoretical basis for the influence spread model considering the existence of competitive and cooperative information.(4)A seed node selection method based on multiplex network representation learning methods is proposed.Based on the previous research,a reversed random walk centrality based on multiplex network representation learning is proposed.The proposed method first uses multiplex network representation learning algorithms to learn the representations of nodes in different layers,and then calculates the similarity weights between nodes in each layer;runs the reversed random walk based on the similarity weight,then count the number of times each node is visited;then realize the selection of top-k seed nodes.In terms of the influence spread model,not only the single influence spread process is considered,but also the spread model with competitive and cooperative information is studied.Under different influence spread models,experiments on multiple datasets verify the effectiveness of the proposed method.