Research And Implementation Of Large-scale Social Information Network Fusion System

The booming development of the mobile Internet makes it easy for people to establish relationships,so there are highly dispersed online social networks around the world.Such a large amount of online data can be used to characterize the relationship between entities,making social network analysis a research hotspot.With the deepening of research,network analysis has developed from the early analysis of single structure to the joint mining of multiple network structures and multi-source information.Social network fusion analysis is to integrate multiple sources,multiple relationship types and heterogeneous information,and use a unified analysis framework to perform fusion analysis and achieve collaborative mining tasks.Based on this background,this paoer has conducted research on a fusion analysis system for large-scale information networks to solve the problem of insufficient accuracy of the description of a single data source.The research content is mainly divided into the following three points:1.We proposed a fusion modeling method for academic social networks based on multiple types of data.This research is the first to introduce a new academic social network based on acknowledgment text data.Then the same entities are modeled by co-authorship network based on the semi-structured data,that is,multiple networks are used to model different social relationships of the same entity.This research solved the network optimization and alignment problems of multi-type data forming a multi-layer network,and finally obtained an academic social multi-layer network.2.We proposed a network structure fusion algorithm SFMN based on semi-supervised learning.The fusion algorithm is divided into two parts,feature extraction and prediction model.Firstly,perform structure-level network feature extraction on the multiplex network,remove the redundant information and identify the complementary information in the network.Then,train the gradient enhancement decision tree to improve the ability to retain complementary information.Finally,the multi-layer network information is constructed as a single-layer network to achieve the purpose of reducing the number of network layers and saving computational overhead.The SFMN model shows better performance than other models in link prediction and community detection tasks.The research also proposed a distributed fusion algorithm based on Spark(PSFMN)to achieve large-scale network calculation and analysis.3.We designed and implemented a large-scale information network fusion analysis prototype system.Based on distributed frameworks,the system provided the function of importing data from multiple sources into the system to construct a network structure and complete a variety of network analysis tasks.Combining mainstream front-end and back-end frameworks and non-relational database technologies,it provided users with friendly interaction methods and interfaces to realize large-scale network integration analysis and mining.