| Complex systems in the real world can be abstracted into complex networks using the method of network science and one of the important features of complex networks is the existence of community structure.Mining the community structure in a complex network through the community detection algorithm can deeply understand the features of the network structure and explore the deeper connections between nodes on the network.At present,the research results of community detection in complex networks have been widely used in networks of different fields,such as social networks,cooperative networks,biological networks and transportation networks,etc.Node influence is an important feature of nodes in the network,and the research on the influence of nodes in the network can promote the development of other network science fields.This paper proposes a calculation method of node influence,and then improves two community detection algorithms based on this method and applies them to the sub-project of Key Laboratory of Media Convergence Technology and Communication in Gansu province.The main research contents are as follows:(1)The steady-state probability obtained after random walk with restart of nodes on the network can well represent the correlation between nodes,but due to its high complexity,it cannot be applied to large-scale networks.Considering the algorithm complexity and algorithm performance comprehensively,this paper proposes a node influence calculation method NILR(Node Influence based Local Random Walk with Restart): the walking range of a node is limited to its first-order neighborhood subgraph,and the probability distribution of a node in steady-state is used to represent the mutual influence of neighbor nodes on it.And then,the influence of a node in the network is represented by the sum of the mutual influence of the node on its neighbors.(2)In order to solve the instability problem and improve the performance of the label propagation algorithm(LPA),this paper proposes a NILR-based label propagation community detection algorithm NILR-LPA.The algorithm uses the method NILR introduced above for precomputation and serialize the update order of nodes in ascending order of node influence.And in the label propagation process,a parameter is set to select the neighbor range of the label propagation,the label selection is guided by the sum of mutual influence of the node labels.At the same time,a post-processing method is also employed to improve the accuracy of the algorithm.Finally,the effectiveness of the algorithm is verified by comparative experiments in artificial networks and real networks.(3)Selecting more representative seed nodes and more accurate community expansion strategies are the research focus of seed expansion algorithm.In this paper,a NILR-based seed extension community detection algorithm NILR-SE is proposed.The algorithm first exploits the node influence calculated by NILR algorithm to select a set of seed nodes with a wide coverage,then uses the mutual influence between nodes to obtain the similarity between nodes and employs it to guide the community expansion to achieve the initial community structure.Finally,the labels are initialized according to the initial community structure,and the modified NILR-LPA algorithm is used to merge the communities,which improves the quality of the community structure detected by the algorithm.Compared with NILR-LPA,NILR-SE does not need to adjust parameters,and the label convergence speed is accelerated.In some real network,better experiments results can be achieved.(4)Apply the proposed NILR-SE algorithm to personalized recommendation for news content.According to the user’s reading data,the user preference network is constructed,and the NILR-SE algorithm is used to detect user groups with similar interests,a personalized recommendation mechanism is proposed according to the group features.The community structure results can also be used as the data basis for other research tasks in the project. |
