CN-119887423-B - Social network viewpoint consistency method and device based on multi-agent reinforcement learning

CN119887423BCN 119887423 BCN119887423 BCN 119887423BCN-119887423-B

Abstract

The invention designs a social network viewpoint consistency method and device based on multi-agent reinforcement learning, comprising the following steps: a critical neighbor selection strategy based on a limited confidence interval, and an external media learning strategy based on reinforcement learning. Confidence intervals among public media are adaptively adjusted according to the harmony degree, and learning optimization strategies are conducted through local public intelligent agent information. The confidence interval between the public intelligent agents is set to be a fixed value, a neighbor set of the intelligent agent i at a certain moment is obtained, the intelligent agent i selects a neighbor with the largest state difference value with the intelligent agent i, and then selects a neighbor with the smallest state difference value with the intelligent agent i. And then, carrying out communication and state evolution according to the screened neighbor information and the external media information in the dynamic confidence interval. The media-public consistency protocol can adapt to a large-scale high-density social network scene and has the effect of enhancing the viewpoint consistency.

Inventors

XIE GUANGQIANG
Xiao Zhuanghong
LI YANG

Assignees

广东工业大学

Dates

Publication Date: 20260512
Application Date: 20241220

Claims (8)

1. A social network view consistency method based on multi-agent reinforcement learning is characterized by comprising the following steps: Under the viewpoint dynamics model of the distributed discrete time continuous state space, the intelligent agent scans all public neighbor intelligent agents at any moment and stores neighbor information in a limited confidence interval; The agent finds out an agent with the largest difference value with the self state in the neighbor information, and finds out an agent with the smallest difference value with the self state; The intelligent agent calculates harmony according to any neighbor information in all the limited confidence intervals; Judging that the public intelligent body can accept external media information by using reinforcement learning; The public agent exchanges information with the screened agent information, control input is obtained through a consistency protocol, and the state information of the public agent is updated through the control input quantity, so that the information of all agent neighbors and external media information can be effectively combined, and the consistency of social network views is achieved; the agent information comprises information of public agents and external media agents; the method for judging that the public intelligent body can accept the external media information by utilizing reinforcement learning comprises the following steps: the preset media quantity is fixed to be 1, the public intelligent agent randomly selects part of public information to perform reinforcement learning every time the public intelligent agent iterates, and after training, randomly selects a value as new external media information according to the normal probability distribution of the information If (if) Then it is explained that the public agent can accept the external media information ; Viewpoint information of public intelligent agent i at time k; HD is harmony; The harmony HD is an index for measuring the degree of the difference in views between the public intelligent agent i and all the neighbors thereof, and is defined as follows: ; Where n is the number of the public, And The viewpoint information of the public agent i and the public agent j are respectively indicated.
2. The multi-agent reinforcement learning-based social networking view consistency method of claim 1, wherein: The multi-agent system is a topology structure depending on states, the viewpoint information of the agent indicates the state information thereof, only one-dimensional viewpoint information of the agent is considered, the viewpoint of the agent at the next moment is determined by the viewpoints of neighbors and external media information in a confidence interval thereof, and the state information at the next moment is expressed as follows: ; Wherein, the The state information of the intelligent agent i at the moment k; is external media information.
3. The multi-agent reinforcement learning-based social networking view consistency method of claim 1, wherein: neighbor information Is defined as for the public agents i and j if - R is the confidence interval length of the agent, and then the agents i and j are neighbors; But can accept or not accept the external media information for the public intelligent body i Depending on - Whether it is true or not, Indicating the harmony of the intelligent agent i, and if so, indicating that the intelligent agent i can accept the external media information.
4. The multi-agent reinforcement learning-based social networking view consistency method of claim 1, wherein: The consistency protocol is as follows: ; Wherein, the , For the adjustment factor, a number greater than 0 and less than 1, Indicating whether the public agent i is connected to an external medium.
5. The multi-agent reinforcement learning-based social networking view consistency method of claim 4, wherein: the communication topology connection of the public intelligent body and the external medium is also time-varying and is used for If indicated, if And 1 if not, 0 if not, and HD is harmony.
6. The multi-agent reinforcement learning-based social networking view consistency method of claim 1, wherein: The conditions for the public agent to reach agreement are: for public agents i and j, if the condition is satisfied - And if the number is equal to 0, the task of consistent views is completed.
7. A social networking point of view consistent electronic device, comprising: a storage medium storing a computer program; A processing unit, in data exchange with the storage medium, for executing the computer program by the processing unit when performing the viewpoint consistency processing, to perform the steps of the social network viewpoint consistency method based on multi-agent reinforcement learning as set forth in any one of claims 1 to 6.
8. A readable storage medium, characterized by: The readable storage medium has a computer program stored therein; The computer program, when run, performs the steps of the multi-agent reinforcement learning based social network perspective consistency method of any of claims 1-6.

Description

Social network viewpoint consistency method and device based on multi-agent reinforcement learning Technical Field The disclosure relates to the field of multi-agent system consistency, in particular to a social network viewpoint consistency method and device based on multi-agent reinforcement learning. Background The opinion is widely present in public life as a form of expression of public opinion. With the rapid development of wireless communication networks and internet technologies, sounding and commenting on daily activities of the public in a network community are gradually performed by using an internet platform. In the network community, the public expresses self-views by utilizing Internet platforms such as WeChat, microblog, QQ and the like to publish individual views. The appearance of multiple views and the continuous interaction of views among individuals make the group views gradually show different phenomena of consistency, dispersion, polarization and the like. Currently, one of the core problems of social network public opinion management is "how to agree on public opinion". The multi-agent system is a novel distributed control system, and compared with the traditional centralized control system of a central controller, the distributed control multi-agent system has the advantages of being powerful in system function, easy to expand, high in cost performance, high in reliability, high in flexibility and strong in robustness. In the existing multi-agent viewpoint consistency enhancement method, most of the methods are combined based on optimization interaction rules, a leader following method, a trust mechanism, external pressure and the like, however, in a social network scene with uneven density distribution and large-scale high density, most of the algorithms are too slow to converge in the scene under a large-scale intelligent system structure, are not applicable, and have no common relevance to how to realize consistency enhancement by using external media information. Disclosure of Invention The method aims to solve the problem that an existing consistency algorithm is too slow to converge under a large-scale intelligent system structure, provides the multi-agent reinforcement learning-based media-public social network viewpoint consistency, utilizes external media information and optimizes a key neighbor selection strategy based on a limited confidence interval, selects more valuable neighbors for communication, reduces communication cost and improves convergence speed. According to the social network viewpoint consistency method based on multi-agent reinforcement learning, media predicts public group viewpoint preference through reinforcement learning, and the public agents comprehensively consider external media information and key neighbor information to update and evolve states, wherein the method comprises the following specific steps: A social network perspective consistency method based on multi-agent reinforcement learning, comprising: Under the viewpoint dynamics model of the distributed discrete time continuous state space, the intelligent agent scans all public neighbor intelligent agents at any moment and stores neighbor information in a limited confidence interval; The agent finds out an agent with the largest difference value with the self state in the neighbor information, and finds out an agent with the smallest difference value with the self state; The intelligent agent calculates harmony according to any neighbor information in all the limited confidence intervals; Judging that the public intelligent body can accept external media information by using reinforcement learning; The public agent exchanges information with the screened agent information, control input is obtained through a consistency protocol, and the state information of the public agent is updated through the control input quantity, so that the information of all agent neighbors and external media information can be effectively combined, and the consistency of social network views is achieved; the agent information includes information of public agents and external media agents. The method for judging that the public intelligent body can accept the external media information by utilizing reinforcement learning comprises the following steps: the preset media quantity is fixed to be 1, the public intelligent agent randomly selects part of public information to perform reinforcement learning every time the public intelligent agent iterates, and after training, randomly selects a value as new external media information according to the normal probability distribution of the information If (if)Then it is explained that the public agent can accept the external media information;And (5) viewpoint information of the public intelligent agent i at the moment k. The multi-agent system is a topology structure depending on states, the viewpoint information of the agent indicates the state information thereof, only