Research On Multi-domain Oriented Heterogeneous Relational Data Transfer Learning Bounds

With the rapid development of the Internet,the number of Internet users in China is increasing.Netizens are active in various network platforms,generating a large amount of network data.Using these data to monitor public opinion and maintain network order.The data is more diverse,and it is difficult to meet the conditions that traditional machine learning requires sufficient tagged data.Transfer learning can cope with this dilemma by using the already tagged data in other fields to transfer knowledge and assist model training.In this paper,we study the transfer learning boundaries for heterogeneous relational data in many fields.Diverse data leads to heterogeneous domains,and most of the reality is relational data.The study of transfer learning limits provides a theoretical basis for the transfer learning effect.Transfering knowledge from multiple domains can be used to model effects.In this paper,we first summarize two kinds of heterogeneous domain adjust methods,and propose the transfer complexity.Then the transfer complexity is integrated into the field distance,then the single source domain transfer learning bounds based on heterogeneous relational data is deduced,which is extended to the multiple-source domain transfer learning bounds,and finally generalizes the transfer learning bounds and carries on the theoretical characteristic analysis.In this paper,the application of transfer learning in the field of public opinion role identification.Firstly,the method of calculating the distance of experience domain is given,then the role identification transfer model based on Markov logic network and the role identification transfer model based on Bayesian logic network are established respectively.This paper selects four network platform’s public opinion data,first calculates the distance between four domains,then establishes the role recognition transfer learning model,observes the change of the transfer learning boundary in the case of transforming the source domain or changing the correlation parameter,and verifies the rationality of the theoretical research.