Research On Semantic Inference Of Social Relationships Based On Human Behavioral Data

In recent years,social relations mining has attracted more and more attention from academic and industrial circles.Researchers hope to restore the original social network from social data.One of the most important researches is to conduct specific semantics of social relations.Understanding the semantics of social relationships can help to grasp the evolution of the microscopic dynamic structure of social networks.In practical applications,it can be used in word-of-mouth marketing in the business sector,and friend recommendation service on social platforms.In the current study,the social networks studied in most jobs only contain a single type of social relationship,such as friend relationships,manager-employee relationships,or support relationships,and different from real social networks that actually contain multiple social relationships.At the same time,in most of the work,researchers only focus on how to mine the semantics of social relationships in online social data,and ignore the research on social data generated by people's interaction activities outside the Internet space.Therefore,in the research of this paper,we will study the real behavior dataset that integrates a variety of social relationships and explore a feasible social relations semantic inference model.In the data set studied in this paper,contains a colleague relationship and romantic partnership,in order to distinguish these two different social relationships from other social relationships at the same time,this paper first analyzes the relationship between colleagues,relationships,and other types of relationship in interactive behavior differences.From the analysis results,it can be seen that different types of social relations difference generally in frequency,diversity,type and time attribute of interaction behaviors.Then,based on the theory of balance in social network theory,the concept of network structure embeddedness and the dispersion of network structure,the characteristics of various social relationships in the network structure are analyzed.In the study of semantic inference model of social relationships,this paper designs a model based on Boosting method that can mine the social relationship semantics in behavioral data.In order to further improve the performance of the model,this paper proposes an algorithm EIA that can evaluate the influence of the learning error on the learning algorithm of the Boosting method.According to the classical SAMME algorithm and EIA algorithm,the model EIASAMME proposed in this paper is obtained.In the last part of this paper,at first,we test the influence of the parameter ? on the performance of the EIA-SAMME algorithm.In the comparison experiment,we compare the impact of different learning error network influence evaluation methods on SAMME algorithm and the performance of EIA-SAMME algorithm,Logistic algorithm and SVM algorithm in semantic inference of social relationships.The experimental results show that EIA-SAMME algorithm performs best in the ability of semantic inference of social relationships.At the same time,during the experiment process,we verified the role of the interaction behavior characteristics and the network structure characteristics in the inference of various social relationships.The results show that when the interaction behavior characteristics can not describe the differences of some social relationships well,the network structure characteristics can well compensate for this deficiency and improve the performance of the social relationship semantic inference model.