A confidence interval-based multi-topic opinion evolution method

Abstract

The application provides a multi-topic opinion evolution method based on a confidence interval, including: an agent i collects and stores neighbor information N i (k); the agent i performs distance calculation on the collected neighbor information and the opinion of the agent i, so as to classify the neighbors and obtain neighbor categories; according to the neighbor categories, neighbor N b and neighbor N c with large opinion differences are obtained; the trust of neighbor N b and neighbor N c is discounted, and the discounted trust degree and the discounted trust degree matrix are obtained; a feedback matrix is derived according to the neighbor categories, a diagonal matrix of the allocable trust degree is obtained through the feedback matrix and the trust matrix, and the element d ii on the diagonal line of the diagonal matrix represents the trust degree that can be redistributed by the agent i; the trust degree is redistributed through the discounted trust matrix; and the multi-topic opinion evolution is completed through a dynamic formula. The method can realize stable consensus on multiple topics with logical dependency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of social network opinion evolution, and in particular to a multi-topic opinion evolution method based on confidence interval. BACKGROUND

[0002] In the digital age of the 21st century, social network platforms such as Douyin, Baidu Post Bar and Weibo have become an essential part of daily life and a core channel for information transmission and public opinion building. These platforms enable users to freely share and express their views and emotions on various events, products or opinions. The posts published by users on social networks not only reflect their individual positions, but also can have a significant impact on the attitudes and behaviors of other users. Therefore, exploring the trend of public opinion in social networks, i.e., how opinions are formed and evolved when the public discusses a particular issue, has become an important topic in social science research and information technology.

[0003] Theoretically, opinion dynamics is to describe the evolution of opinions among a group of interacting individuals. Based on the fact that each person holds their own opinion and exchanges opinions with each other, social networks can be modeled. "Consensus", "polarization" and "splitting" are the main features of the final stage of group opinion evolution. By modeling social networks, the direction of public opinion can be analyzed, which helps to better guide public opinion, respond to network public opinion emergencies, and prevent the occurrence and spread of rumors.

[0004] Existing social network opinion evolution models mostly focus on the study of a single topic, but such methods are not suitable for discussing a group of topics with logical dependencies. When discussing a group of topics with logical dependencies, the affirmation of one topic is often influenced by the mixture of other topics. Moreover, if there are opinion extremists in the discussion group, the opinions of others will be severely disturbed.

[0005] Therefore, the key problem is how to establish the dependency relationship between topics in a weighted directed topology at discrete time and reduce the disturbance caused by extreme opinions. SUMMARY

[0006] The purpose of the present application is to overcome the problem that in the existing opinion evolution model, the public is disturbed by extreme opinions in multi-topic opinion evolution, leading to a serious deviation from their own views. The present application provides a multi-topic opinion evolution method based on confidence interval, which classifies the neighbors who exchange opinions, reduces the trust of neighbors with large opinion differences, and enhances the trust of neighbors with similar opinions.

[0007] The specific technical solution of the present application is:

[0008] A multi-topic opinion evolution method based on confidence interval, comprising:

[0009] In the discrete-time model fixed topology, at time k, the agent i collects and stores the neighbor information N i (k);

[0010] The agent i calculates the distance between the collected neighbor information and its own opinion to classify the neighbors and obtain neighbor categories;

[0011] According to the neighbor categories, the neighbors N b with large opinion differences are obtained c , and the neighbors N b and N c are discounted in trust to obtain the discounted trust degree and the discounted trust degree matrix;

[0012] According to the neighbor categories, a feedback matrix is derived, and a diagonal matrix of allocatable trust degrees is obtained through the feedback matrix and the trust matrix, wherein the elements d ii on the diagonal line of the diagonal matrix represent the trust degrees that can be redistributed by the agent i, and the trust degree is redistributed to the discounted trust matrix;

[0013] The opinion information of the agent is updated through a dynamic formula, and the multi-topic opinion evolution is completed.

[0014] The agent i calculates the distance between the collected neighbor information and its own opinion to classify the neighbors and obtain neighbor categories, including:

[0015] The agent i calculates the distance between the collected neighbor information and its own opinion, and each agent has a confidence interval wherein and are respectively referred to as the lower limit and the upper limit of the confidence degree of the agent i;

[0016] According to whether the opinions of different topics of the neighbors are located in the confidence interval of the agent itself, the classification is performed.

[0017] The classification according to whether the opinions of different topics of the neighbors are located in the confidence interval of the agent itself, including:

[0018] The A-class neighbor N a is a neighbor whose distances of all topics opinions are within the confidence interval of the agent itself;

[0019] The B-class neighbor N b is a neighbor whose distances of some topic opinions are within the confidence interval of the agent itself, and distances of some other topic opinions are not within the confidence interval of the agent itself;

[0020] The C-class neighbor N cis all the opinion distance of the topic of the neighbor outside the confidence interval of itself.

[0021] The neighbor N b and the neighbor N c are discounted in trust, including:

[0022] The coefficients a and b are set, and a> b e (0, 1), respectively corresponding to the neighbor N b and the neighbor N c , which are discounted in trust, that is, or Wherein is the discounted trust degree; w ij is the original trust degree.

[0023] The feedback matrix is derived according to the neighbor category, including:

[0024] The feedback matrix F e R n×n is derived according to the neighbor category:

[0025]

[0026] Wherein f ij represents the element of the i-th row and the j-th column in the feedback matrix, that is, the feedback coefficient of individual i to neighbor j.

[0027] The diagonal matrix of the allocable trust degree is obtained by the feedback matrix and the trust matrix, and the element d ii on the diagonal line of the diagonal matrix represents the trust degree that can be redistributed by the intelligent agent i, and the trust degree is redistributed after the discounted trust matrix is discounted, including:

[0028] After obtaining the feedback matrix, the diagonal matrix D = diag (W x F T ) of the allocable trust degree is obtained by the operation of the feedback matrix and the trust matrix, and the element d ii on the diagonal line represents the trust degree that can be redistributed by the intelligent agent i, and finally the trust degree is redistributed after the discounted trust matrix is discounted, that is,

[0029] Wherein, W is the trust matrix; F T is the transpose of the feedback matrix.

[0030] The opinion information of the intelligent agent is updated by the dynamic formula, and the multi-topic opinion evolution is completed, including:

[0031] The dynamic formula of the logical dependency relationship between the topics is established, and the opinion iteration evolution is performed by the dynamic formula, the opinion information of the intelligent agent is updated, and the evolution task is completed according to the condition.

[0032] The dynamics formula is:

[0033]

[0034] The condition for completing the evolution task is:

[0035] For the intelligent agent i, if the condition lim k→∞ ||x i (k+1)-x i (k)||=0 is met, the evolution task is completed.

[0036] Wherein, x i (k+1) is the opinion of individual i at the k+1 time; x i (k) is the opinion of individual i at the k time, and when the difference between the two opinions is infinitely close to 0, it indicates that the evolution task is completed.

[0037] Beneficial effects:

[0038] The beneficial effects of the present application at least include:

[0039] The method provided by the present application first collects and stores neighbor information N i (k); and calculates the distance between the collected neighbor information and the self opinion to classify the neighbors and obtain neighbor categories; then, according to the neighbor categories, obtains neighbors N b and N c with large opinion differences; and discounts the trust of the neighbors N b and N c , obtains the discounted trust and the discounted trust matrix; further, derives a feedback matrix according to the neighbor categories, obtains a diagonal matrix of allocatable trust degrees through the feedback matrix and the trust matrix, and the element d ii on the diagonal line of the diagonal matrix represents the trust degree that can be redistributed by the intelligent agent i, and the discounted trust matrix is subjected to trust degree redistribution; finally, the opinion information of the intelligent agent is updated through the dynamics formula, and the multi-topic opinion evolution is completed. The present application classifies the neighbors through the confidence interval, and adjusts the trust degree of the neighbors, so that the trust degree of the neighbors holding extreme opinions is weakened, and the interference of the extreme opinions on the opinion of the intelligent agent itself is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of the method provided by the present application.

[0041] Figure 2 is the simulation result of the multi-topic opinion evolution model with confidence interval in a self-defined network.

[0042] Figure 3Simulation results of the multi-topic opinion evolution model with confidence interval in ER network. DETAILED DESCRIPTION

[0043] The present application is further described below with reference to the accompanying drawings.

[0044] In order to make the present application more detailed, the prior art needs to be described in detail.

[0045] The multi-topic opinion evolution method based on confidence interval proposed in the application updates the trust degree by hierarchically dividing the neighbors through the confidence interval neighbor classification algorithm, such as Figure 1 , the specific steps are as follows:

[0046] Step one: under the fixed topology structure of the discrete time model, at time k, individual i collects and stores neighbor information N i (k);

[0047] Step two: individual i calculates the distance between the collected neighbor information and its own opinion:

[0048] And each individual has a confidence interval Where and are called the lower and upper bounds of the confidence degree of individual i respectively. According to whether the opinions of different topics of the neighbors are within the confidence interval of itself, the neighbors are divided into three categories: A-class N a is the neighbor whose opinion distance of all topics is within the confidence interval of itself; B-class N b is the neighbor whose opinion distance of some topics is within the confidence interval of itself, and whose opinion distance of some topics is not within the confidence interval of itself; and C-class N c is the neighbor whose opinion distance of all topics is outside the confidence interval of itself.

[0049] Step three: after the classification of the neighbors, coefficients α and β are set respectively, and α>β∈(0,1), which respectively correspond to the neighbors N b with large opinion difference and the neighbors N c , and the trust degree of which is discounted. That is or Where is the discounted trust degree.

[0050] Step four: in order to realize the redistribution of the trust degree, it is necessary to derive the feedback matrix F∈R n×n according to the neighbor categories:

[0051]

[0052] Step 5: After obtaining the feedback matrix, we can obtain a diagonal matrix D = diag(W × F) through the operation between the feedback matrix and the trust matrix. T ), its diagonal element d ii This represents the trust level that individual i can redistribute. Finally, the discounted trust matrix is... To redistribute trust, i.e.

[0053]

[0054] Step Six: The updated trust matrix can be used to update individual opinion information by substituting it into the dynamic formula. Specific Implementation Example 1:

[0056] This invention provides an embodiment:

[0057] In the initial state of opinion evolution, people's opinions on the topic are collected and discretized into values ​​between [-1, 1], where [-1, 0) represents opposition and (0, 1] represents agreement. X(k) is the opinion matrix of all individuals at time k. in Let represent the attitude of individual i towards the m-th topic at time k. Simultaneously, a mathematical model of the trust relationships between people is constructed and represented by a trust matrix W, W = [w...]. ij ]∈R n×n It is a weight matrix, whose elements w ij Let be the weight of the edge (i,j), representing the degree of trust that individual i has in individual j. Therefore, for a given matrix W, it uniquely defines a network, denoted as G(W). The neighborhood set of individual i is defined as N. i ={j|w ij ≠0, j≠i}. When j=i, w ij This is considered the degree of adherence to one's own opinions. Furthermore, matrix W is a row-random matrix, meaning that for... In other words All are true. This is achieved through a multi-problem dependency structure matrix. Establish dependencies between topics.

[0058] After setting the initial state, the opinions are iteratively evolved using dynamic formulas:

[0059]

[0060] Where, λ i This represents the sensitivity of individual i to external influences, with values ​​ranging from [0,1], and is typically taken as λ. i =1-w ii That is, self-weight w iiis considered as an indicator of the sensitivity or closeness of the ith individual to interpersonal influence. i is the obsession of the individual, usually taken as i = x i (0), i.e., the initial opinion of the individual. The specific inputs and outputs of the model are as follows:

[0061]

[0062]

[0063] Compared with existing research algorithms such as Figure 2 , 3, the present application has the advantage of classifying neighbors by confidence intervals, reducing the trust of neighbors with very different opinions, and enhancing the trust of neighbors with similar opinions, thereby improving the resistance of individuals to extreme opinions. Specific embodiment 2:

[0065] The present application also provides an embodiment:

[0066] An electronic device, comprising: a storage medium and a processing unit; wherein the storage medium is used to store a computer program, the processing unit exchanges data with the storage medium, and is used to execute the computer program by the processing unit when performing multi-topic opinion evolution, and perform the steps of the confidence interval-based multi-topic opinion evolution method as described in specific embodiment 1. Specific embodiment 3:

[0068] A computer-readable storage medium, the computer-readable storage medium stores a computer program; the computer program, when executed, performs the steps of the confidence interval-based multi-topic opinion evolution method as described in specific embodiment 1.

[0069] In the present application, the computer-readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, device or apparatus. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signal, optical signal or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, transmit or propagate programs for use by or in conjunction with an instruction execution system, device or apparatus. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination thereof.

[0070] The above disclosed are only several specific implementation scenarios of the present application, but the present application is not limited thereto, and any variations that can be thought of by those skilled in the art shall fall within the protection scope of the present application. The above application serial numbers are only for description, and do not represent the advantages and disadvantages of the implementation scenarios.

Claims

1.A method for opinion evolution of multiple topics based on confidence interval, characterized in that, The method comprises the following steps: In the discrete-time model with fixed topology, at time k, agent i collects and stores neighborhood information N i (k); The agent i calculates the distance between the collected neighbor information and its own opinion to classify the neighbors and obtain neighbor categories; According to the neighbor category, a neighbor N with a large difference in opinion is obtained b and the neighbor N c ; and the neighbor N b and the neighbor N c is discounted, to obtain a discounted trust degree and a discounted trust degree matrix The feedback matrix is derived according to the neighbor category, the diagonal matrix of the allocable trust degree is obtained through the feedback matrix and the trust matrix, and the elements d on the diagonal line of the diagonal matrix represent the trust degree that can be redistributed by the intelligent agent i ii The discounted trust matrix is subjected to trust degree redistribution. The agent opinion information is updated through a dynamic formula to complete the multi-topic opinion evolution; The and the neighbor N b And the neighbor N c Carry out trust discount, including: Set the coefficients a, b and a > b ∈ (0, 1), respectively, corresponding to the existence of a larger difference of opinion neighbors N b And the neighbors N c , the trust discount, that is Or Where The discounted trust degree; w ij The original trust degree; The feedback matrix is derived according to the neighbor categories, and the method comprises the following steps: The feedback matrix F e R is derived from the neighbor class n×n : wherein, wherein f ij denotes the element in the feedback matrix in the i-th row and j-th column, i.e. the feedback coefficient of individual i on neighbor j; The diagonal matrix of the allocable trust degree is obtained by the feedback matrix and the trust matrix, and the element d on the diagonal line of the diagonal matrix represents the trust degree that can be redistributed by the intelligent agent i ii The discounted trust matrix is subjected to trust degree redistribution, and the discounted trust matrix represents the trust degree that can be redistributed by the intelligent agent i. After obtaining the feedback matrix, a diagonal matrix D = diag(W x F T ) is obtained by operating the feedback matrix and the trust matrix, wherein the element d ii on the diagonal represents the redistributable trust degree of the intelligent agent i, and finally the trust degree is redistributed to the discounted trust matrix , that is wherein W is a trust matrix; F T is the transpose of the feedback matrix. 2.The confidence interval based multi-topic opinion evolution method of claim 1, wherein, The agent i calculates the distance between the collected neighbor information and its own opinion to classify the neighbors and obtain neighbor categories, and the method comprises the following steps: Agent i performs distance calculation on collected neighbor information and its own opinion, And each agent has a confidence interval - lower i ≤ x j - x i ≤ upper i Where and are called lower and upper confidence bounds of agent i, respectively. The classification is performed according to whether the opinions of the neighbors on different topics are located in the confidence interval of the agent. 3.The confidence interval based multi-topic opinion evolution method of claim 2, wherein, The classification is performed according to whether the opinions of the neighbors on different topics are located in the confidence interval of the agent. Class A neighbors N a is a neighbor whose distance from all topics opinions is within its own confidence interval; Class B neighbors N b There are some neighbors whose distance is within the confidence interval of the self, and some are not. Class C neighbors N c are all neighbors whose opinion distance is outside of their own confidence interval. 4.The confidence interval based multi-topic opinion evolution method of claim 1, wherein, The agent opinion information is updated through a dynamic formula to complete the multi-topic opinion evolution, and the method comprises the following steps: A dynamic formula of logical dependency relationship between topics is established, and the opinion is iteratively evolved through the dynamic formula to update the agent opinion information and complete the evolution task according to conditions. 5.The confidence interval based multi-topic opinion evolution method of claim 4, wherein, The dynamic formula is as follows: where λ i represents the sensitivity of individual i to external influences; u i is the individual's obsession; C is the structure matrix. 6.The confidence interval based multi-topic opinion evolution method of claim 4, wherein, The condition for completing the evolution task is as follows: For agent i, if the condition lim k→∞ ||x i (k+1)-x i (k)|| = 0 is satisfied, the task of evolution is completed. where x i (k+1) is the opinion of individual i at time k+1; x i (k) is the opinion of individual i at time k, and the difference between the two opinions approaches zero as the evolutionary task is completed.

Images