Research On Argumentation Based Multi-agent Collaboration Classification With Privacy Preserving

With the accelerating social development of information and network,data mining and machine learning in distributed environments,especially in the process of classification technology,have revealed a lot of implicit knowledge and created wealth for people,while the process is under the severe ordeal of privacy and information security.At the same time,the interpretation of the classification results has gradually attracted the attention of researchers in the field of intelligent decision-making.Therefore,how to solve the classification problem in the distributed environment under the condition of privacy preserving and give a reasonable explanation to the classification results,becomes one of the most important and challenging frontier topics.This paper is based on the theory of argumentation,focusing on the theme of "Research on Argumentation based Multi-agent Collaboration Classification with Privacy Preserving",and applies knowledge of Multi-Agent system technology,argument game dialogue theory,inductive learning technology,ontology and other related fields comprehensively to analyze the argumentation model of Multi-Agent collaboration classification method,the construction and performance of argument and optimization method for hierarchical data model systematicly.The main research work of this paper includes the following parts.(1)Focusing on how the agents interact with each other under the privacy protection condition,combined with the dialogue game thought in the theory of argumentation,a new Multi-Agent argumentation model A-MACC,which is based on rational argumentation space is established.This model converts the Agents argumentation into pair argument game process by introducing the concept of "pairwise strength",which makes sure that the cooperative classification between Agents can be done orderly and under condition of privacy preserving.With the concept of argument pairwise preference relations and to defeat relations as the foundation,we construct the multi argument game dialogue process in A-MACC,and analyze important properties of Multi-Agent arugmentation.The results of these studies will further improve and extend argumentation theory in the field of artificial intelligence.(2)Focusing on how the argument is produced during the argumentation,the Multi-Agent collaboration classification method Arguing-Prism based on A-MACC model is studied,and the dynamic construction algorithms based on Prism are proposed.This algorithm introduces the Prism classification rule induction learning into the Multi-Agent argumentation,and realizes the automatic construction of the classification argument.Further,the Agent cooperative classification method Arguing-Prism is analyzed,and the main factors that affect the classification performance of the proposed method are analyzed from two aspects of the argument constructing strategy and the inconsistency of the data.A large number of experiments on UCI benchmark datasets show that the proposed method is effective in the comprehensive performance of a variety of classification indexes.These research works will effectively promote the application of argumentation theory in machine learning and data mining.(3)Aiming at the hierarchical data,a domain ontology guided modular classification rule inductive learning algorithm(SATE-Prism)is proposed,which is used to improve the robustness of the Multi-Agent cooperative classification method.This algorithm combines the idea of domain ontology assisted machine learning,and by introducing the related theory of hierarchical data attribute value generalization,the generalization process of attribute values based on semantic attribute value tree SAT is studied.Experiments show that the proposed algorithm can effectively improve the robustness of the Multi-Agent cooperative classification method based on argumentation.This research work promotes the application of ontology methods in the study of hierarchical data classification.