Research On Overlapping Community-based Influence Maximization

In recent years,with the rapid development of Internet Technology,various social networks are available.Interactive behaviors of users produce massive amounts of data,which has good value of research in marketing,advertising,information recommendation and public opinion monitoring.The data brings new opportunities and challenges to social network analysis.Research on influence maximization has become a research hotspot in social networks.Influence maximization aims at mining a seed set under certain propagation model that influences the largest number of nodes in the social network.Community division is the main method of dealing with the large-scale social networks.Most researchers research on overlapping communities in order to reflect the structure of real network more objectively.However,the existing overlapping community partitioning algorithms are mostly unstable.What's worse,the influence of overlapping nodes in the information propagation process is not fully considered,the propagation range of the set of nodes obtained based on the impact maximization scheme of such algorithms is relatively small.The propagation range of nodes obtained by the influence maximization schemes based on this algorithm is relatively small,ignoring the influence of overlapping nodes in the information dissemination process.In order to comprehensively consider to the structure of real social network and improve the accuracy of the influence maximization algorithm,we proposed a multi-label propagation algorithm and an overlapping community-based influence maximization algorithm.Our main work is as follows:(1)In order to solve the problem of non-deterministic propagation process,unstable results of community division and low community quality in the adoption of a random strategy of the COPRA algorithm,this paper proposed a Multi-Label Propagation Algorithm based on the Node Comprehensive Similarity(MLPA-NCS)for community division based on the COPRA algorithm.Firstly,the algorithm updates node order using the descending order of the potential influence of the user node.In this way,the problem of the unstable division of the community results caused by the randomly selected node update order can be addressed.Then,our algorithm uses node synthesis similarity as the order of traversing neighboring nodes when updating node labels.This strategy can fully consider the topic similarity factors and link relationships between nodes,avoiding the unstable problems caused by updating the labels due to random strategies and improving the quality of the generated community.Finally,the experimental results show that our proposed MLPA-NCS algorithm is superior to the COPRA algorithm and existing community partitioning algorithms considering the NMI and Qov indicators.The community division result of this algorithm is stable and reasonable.(2)In order to solve the problem of the overlapping community-based influence maximization,this paper proposed Overlap Factor-based Core Covering Algorithm(OFCCA)and a NICM propagation model on the basis of the above results.The OFCCA algorithm,which is based on the CCA,selects the candidate node sets.The OFCCA algorithm improves the quality of candidate nodes sets in overlapping communities by considering the influence of overlapping factors on information spreading.NICM model calculates the activation probability with the node intimacy,node topic similarity and information acceptance.This model that simulates the information propagation process can improve the accuracy of the target core node set.Experimental results show that the OFCCA algorithm and the NICM model have advantages in the range and time efficiency of the set of target core nodes mining in overlapping communities.