A method and system for positive guidance of public opinion based on dual learning

Abstract

The present invention discloses a method and system for positively guiding public opinion based on dual learning. The main task is to use dual learning to perform positively guiding public opinion, which mainly includes: dividing the social network that needs public opinion guidance into communities through the GN algorithm, calculating the emotional tendency of the nodes in the network according to the emotional classification, and obtaining the emotional polarity of each node; combining the PageRank algorithm with the emotional polarity, discovering the opinion leaders in the community, and distinguishing them into active guides and passive receivers; constructing a dual learning model based on the dual learning of machine learning. Setting dual tasks, the active guide assigns guidance weights to guide the passive receivers, and the state changes of the passive receivers and the state changes of the network are fed back to the active guide. Designing the characteristic parameters of the training model, calculating the learning convergence conditions, and repeating this process; using opinion dynamics to guide the remaining nodes in the community, and then calculating the proportion of nodes in the positive state, judging the public opinion guidance results, and completing the positive guidance of public opinion.

Description

Technical Field

[0001] The present invention belongs to the field of social networks, and in particular relates to a method and system for positively guiding public opinion based on dual learning. Background Art

[0002] With the development of Internet technology, social networks have become an important platform for people to obtain information and communicate with each other, and the use of social media is one of the most popular online activities. Compared with traditional public opinion, online public opinion has become one of the important ways for people to express their opinions and understand each other. In the process of forming online public opinion, netizens' views will change in mutual interaction and integration. When a large number of netizens hold negative views, it may have a huge impact on economic development and social stability. Moreover, at this stage, my country is in a period of social transformation, and the monitoring and supervision of online information has just begun. Some legal regulations on the Internet are not yet sound enough. Therefore, the Internet has become a channel for many people to express their demands, and online social media has gradually developed from a source of information passively obtained by netizens to a public platform for expressing personal opinions, discussing public topics, and collective rights protection. Online communities have become a gathering place for different information and views. Therefore, online public opinion has developed rapidly in such an environment.

[0003] With the diversification of social networks and the rapid development of the Internet, online public opinion is closely related to our lives, and the Internet has become the primary battlefield for the dissemination and diffusion of public opinion. However, the Internet has the characteristics of randomness, virtuality, and concealment, and the existing network platforms also have the characteristics of civilianization, disorganization, and non-standardization, which makes online public opinion, especially negative public opinion, have a serious impact once it breaks out. The wanton growth of these negative public opinions will have a great negative impact on social stability and national security. After manual guidance by government departments, negative public opinion can be well controlled and even become positive public opinion, which can make people's lives more positive. However, manual guidance may miss the best time to guide public opinion. Therefore, how to use computer technology to automatically and timely guide public opinion events to reduce harm to society is currently a challenging research.

[0004] Early public opinion guidance algorithms mainly used node topology information (in-degree, out-degree, PageRank value, structural holes, etc.) to select guided nodes. However, these nodes have the following three significant shortcomings in public opinion guidance:

[0005] (1) When a single metric is used, such as using in-degree or out-degree to select nodes, the selected nodes are not representative and are not necessarily the most influential nodes in the network.

[0006] (2) Using static methods, which are difficult to adapt to dynamically changing networks and are not adaptive, while machine learning methods can effectively handle network changes and make better decisions.

[0007] (3) The strategies adopted in different networks cannot be migrated. The strategies adopted in small-scale networks or dense networks may not be applicable to large-scale networks or sparse networks. However, by using graph embedding technology, we can migrate the strategies adopted in small-scale networks to large-scale networks, which has high scalability. Summary of the invention

[0008] In order to solve the technical problems existing in the background technology, the present invention aims to provide a method and system for positively guiding public opinion based on dual learning.

[0009] In order to solve the technical problem, the technical solution of the present invention is:

[0010] A method for positively guiding public opinion based on dual learning, the method comprising:

[0011] Construct a social network graph and preprocess it to obtain community division results and user sentiment polarity;

[0012] Using PageRank algorithm and combining users’ sentiment polarity, we search for opinion leaders in the community among the community division results.

[0013] Establish a dual public opinion positive guidance model to conduct positive public opinion guidance for community opinion leaders.

[0014] Furthermore, after searching and obtaining the opinion leaders of the community in the community division results, the method further includes: distinguishing the active guides and the passive receivers in the opinion leader area of ​​the community.

[0015] Furthermore, crawl the social network data, analyze the social network data through the preset networkX library, and build a social network graph.

[0016] Furthermore, the Girvan Newman algorithm is used to divide the social network graph into communities and obtain the community division results; wherein, the community division uses modularity as an update indicator.

[0017] Furthermore, the user features in the social network graph are used to input the user features into a preset feedforward neural network to obtain the user's emotional polarity.

[0018] Furthermore, the opinion leaders of the community in the community division results obtained by the search specifically include:

[0019] The PageRank algorithm in the preset networkX library is used to calculate the nodes in the social network graph to obtain the PageRank value of each node in the social network graph;

[0020] The PageRank value of each node in the social network graph is combined with the above-obtained user sentiment polarity to obtain the user influence score;

[0021] The user influence scores are sorted to obtain the opinion leaders of the community in the community division results.

[0022] Furthermore, according to the emotion polarity of the users, the opinion leaders of the community in the community division result are distinguished to obtain the active guides and passive receivers in the opinion leader area of ​​the community.

[0023] Furthermore, based on the dual learning method in machine learning, a dual public opinion positive guidance model is constructed:

[0024] Firstly, a dual task is set up, and a guidance weight is assigned to the active guide to guide the passive receiver; the emotional polarity state change information of the passive receiver and the state change information of the network are fed back to the active guide; the characteristic parameters of the dual public opinion positive guidance model are given to perform public opinion positive guidance. If the emotional polarity state change information of the passive receiver and the state change information of the network do not become positive, this process is repeated until the emotional polarity state change information of the passive receiver and the state change information of the network become positive, and the public opinion positive guidance is successful.

[0025] Furthermore, opinion dynamics and the dual public opinion positive guidance model are used to guide the nodes around the passive receivers, so that the state of the entire social network graph becomes positive, and the state of the social network graph is judged according to the set proportion of positive state nodes, thereby achieving positive guidance of public opinion.

[0026] A public opinion positive guidance system based on dual learning, the system comprising:

[0027] one or more processors;

[0028] A memory for storing one or more programs;

[0029] When the one or more programs are executed by the one or more processors, the one or more processors execute any of the above-described methods for positively guiding community public opinion based on dual learning.

[0030] Compared with the prior art, the advantages of the present invention are:

[0031] The present invention discloses a novel method for positively guiding public opinion based on dual learning. The main task is to use dual learning to perform positively guiding public opinion, which mainly includes: dividing the social network that needs public opinion guidance into communities through the GN algorithm, calculating the emotional tendency of the nodes in the network according to the emotional classification, and obtaining the emotional polarity of each node; combining the PageRank algorithm with the emotional polarity, discovering the opinion leaders in the community, and distinguishing them into active guides and passive receivers; constructing a dual learning model based on the dual learning of machine learning. Setting dual tasks, the active guide assigns guidance weights to guide the passive receivers, and the state changes of the passive receivers and the state changes of the network are fed back to the active guide. Designing the characteristic parameters of the training model, calculating the learning convergence conditions, and repeating this process; using opinion dynamics to guide the remaining nodes in the community, and then calculating the proportion of nodes in the positive state, judging the public opinion guidance results, and completing the positive guidance of public opinion. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 , a system flow chart of a method for positively guiding public opinion based on dual learning according to the present invention. DETAILED DESCRIPTION

[0033] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0034] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0035] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0036] Embodiment 1:

[0037] A method for positively guiding public opinion based on dual learning, the method comprising:

[0038] Construct a social network graph and preprocess it to obtain community division results and user sentiment polarity;

[0039] Using PageRank algorithm and combining users’ sentiment polarity, we search for opinion leaders in the community among the community division results.

[0040] Establish a dual public opinion positive guidance model to conduct positive public opinion guidance for community opinion leaders.

[0041] Furthermore, after searching and obtaining the opinion leaders of the community in the community division results, the method further includes: distinguishing the active guides and the passive receivers in the opinion leader area of ​​the community.

[0042] Furthermore, crawl the social network data, analyze the social network data through the preset networkX library, and build a social network graph.

[0043] Furthermore, the Girvan Newman algorithm is used to divide the social network graph into communities and obtain the community division results; wherein, the community division uses modularity as an update indicator.

[0044] Furthermore, the user features in the social network graph are used to input the user features into a preset feedforward neural network to obtain the user's emotional polarity.

[0045] Furthermore, the opinion leaders of the community in the community division results obtained by the search specifically include:

[0046] The PageRank algorithm in the preset networkX library is used to calculate the nodes in the social network graph to obtain the PageRank value of each node in the social network graph;

[0047] The PageRank value of each node in the social network graph is combined with the above-obtained user sentiment polarity to obtain the user influence score;

[0048] The user influence scores are sorted to obtain the opinion leaders of the community in the community division results.

[0049] Furthermore, according to the emotion polarity of the users, the opinion leaders of the community in the community division result are distinguished to obtain the active guides and passive receivers in the opinion leader area of ​​the community.

[0050] Furthermore, based on the dual learning method in machine learning, a dual public opinion positive guidance model is constructed:

[0051] Firstly, a dual task is set up, and a guidance weight is assigned to the active guide to guide the passive receiver; the emotional polarity state change information of the passive receiver and the state change information of the network are fed back to the active guide; the characteristic parameters of the dual public opinion positive guidance model are given to perform public opinion positive guidance. If the emotional polarity state change information of the passive receiver and the state change information of the network do not become positive, this process is repeated until the emotional polarity state change information of the passive receiver and the state change information of the network become positive, and the public opinion positive guidance is successful.

[0052] Furthermore, opinion dynamics and the dual public opinion positive guidance model are used to guide the nodes around the passive receivers, so that the state of the entire social network graph becomes positive, and the state of the social network graph is judged according to the set proportion of positive state nodes, thereby achieving positive guidance of public opinion.

[0053] A public opinion positive guidance system based on dual learning, the system comprising:

[0054] one or more processors;

[0055] A memory for storing one or more programs;

[0056] When the one or more programs are executed by the one or more processors, the one or more processors execute any of the above-described methods for positively guiding community public opinion based on dual learning.

[0057] Embodiment 2:

[0058] like Figure 1 As shown, the purpose of the present invention is to provide a method and system for guiding public opinion positively based on dual learning, which can reduce the burden of manual public opinion guidance and automatically and timely guide negative public opinion on the Internet. The overall process of the method is as follows Figure 1 As shown, the specific steps include:

[0059] Step S1: The GN algorithm divides the social network that needs public opinion guidance into communities, calculates the emotional tendency of the nodes in the network according to the emotional classification, and obtains the emotional polarity of each node.

[0060] Step S1.1: Divide the public opinion guidance community.

[0061] The social network dataset is processed using the Python graph analysis library NetworkX, and the Girvan Newman algorithm in the community partitioning module is used to partition the network into k communities C = {c1, c2, ..., c k The currently popular modularity increment ΔQ is used as the update indicator for community division, which is defined as:

[0062] Where m is the total number of edges in the network, and c represents the emotional community that node i moves into. s is the total weight of the edge between node i and the internal nodes of community c. o (c) is represented by the weight of the edge between community c and its external nodes, k i Represents the degree of node i.

[0063] Step S1.2: Calculate the user emotion polarity.

[0064] The user's personal information, published articles, comments, etc. are used as user features to judge the user's emotional polarity. First, various text features are embedded in words, then vectorized and input into the neural network to judge the user's emotional polarity.

[0065]

[0066] Among them, senti(u) represents the sentiment polarity of user u, V(u) 个人信息 represents the vectorized representation of user u’s personal information, f(·) represents the neural network, Represents a vector concatenation operation.

[0067] Step S2: By combining the PageRank algorithm with sentiment polarity, opinion leaders in the community are discovered and divided into active leaders and passive receivers.

[0068] Step S2.1: Calculate the PR value;

[0069] Use NetworkX to get the PageRank value of each node based on the PageRank algorithm in the web page sorting algorithm.

[0070] Step S2.2: Calculate the user influence score, i.e., the influence judgment index;

[0071] The influence of the node in the process of public opinion dissemination is calculated based on the sentiment polarity of each node obtained in S1-2 and the PageRank value obtained in S2-1.

[0072] Influence(u)=|senti(u)*σ(PageRank(u))|;

[0073] Where Influence(u) represents the influence score of node u, PageRank(u) represents the PageRank value of the node, and σ(·) represents the Sigmoid function.

[0074] Step S2.3: Differentiate into active guides and passive receivers;

[0075] In each community divided by S1.1, the node with the greatest influence is regarded as the opinion leader of the community, and then divided into active guides and passive receivers according to the emotional polarity of the opinion leaders.

[0076] Step S3: According to the dual learning method in machine learning, a dual public opinion positive guidance model is constructed, and the dual public opinion positive guidance model includes: a learning model and a dual model.

[0077] Firstly, a dual task is set up, and a guidance weight is assigned to the active guide to guide the passive receiver; the emotional polarity state change information of the passive receiver and the state change information of the network are fed back to the active guide; the characteristic parameters of the dual public opinion positive guidance model are given to perform public opinion positive guidance. If the emotional polarity state change information of the passive receiver and the state change information of the network do not become positive, this process is repeated until the emotional polarity state change information of the passive receiver and the state change information of the network become positive, and the public opinion positive guidance is successful.

[0078] Step S3.1: dual learning assigns guidance weights;

[0079] According to the dual learning of machine learning, a learning model is constructed. According to the network state (feedback), the learning model gives the guidance weights for different passive receivers.

[0080] Step S3.2: obtaining a feedback signal;

[0081] According to the guidance weight given by the learning model in S3.1, the dual model gives the network state (feedback) after guidance as the input of the learning model at the next moment.

[0082] Step S3.3: Construct a dual public opinion positive guidance model;

[0083] Repeat S3.1 and S3.2 until the learning model and dual model parameters converge. The process is as follows:

[0084] W=Model 学习 (State);

[0085] State=Model 对偶 (W);

[0086] Model 学习 and Model 对偶 They represent the learning model and the dual model respectively, W represents the guidance weight, and State represents the network state (feedback).

[0087] Step S4: Use opinion dynamics and the dual public opinion positive guidance model to guide the nodes around the passive receiver, so that the state of the entire social network graph becomes positive, and judge the state of the social network graph according to the set proportion of positive state nodes, so as to achieve positive guidance of public opinion.

[0088] Step S4.1: Use opinion dynamics and the dual public opinion positive guidance model to guide the nodes around the passive receiver;

[0089] Step S4.2: Calculate the positive node rate (the percentage of nodes in the positive state) after the bootstrapping is completed. The formula is as follows:

[0090]

[0091] Where P pos Represents the positive node rate, which is used to measure the quality of the guidance results. pos Indicates the number of positive nodes at the end of the boot, N total Indicates the number of nodes in the entire network.

[0092] Embodiment 3:

[0093] This example is two real social networks collected from the Internet, Sina Weibo and Flickr, and their specific information is shown in Table 1.

[0094] Table 1 Weibo-Flickr network data statistics

[0095] Social Network Number of users Number of relationships Number of posts Sina Weibo 6,474 67,142 66,480 Flickr 4,066 36,236 354,192

[0096] Step 1: Preprocess social network user data, build a social network graph G, obtain user sentiment polarity and user influence, community division, and discover opinion leaders in the community. It includes the following five sub-steps:

[0097] Step 1.1: Build a social network graph;

[0098] The users of the social network are represented as nodes, and the relationships between users (such as friends, followers / fans, etc.) are represented as edges, and an undirected and unweighted graph G = (V, E) is constructed. The users in the two social networks Sina Weibo and Flickr are regarded as nodes V in the graph, and different digital IDs are used to distinguish different users. For example, the users in Weibo correspond to numbers 0 to 6473, and the users in the Flickr network correspond to numbers 0 to 4065. The follow / follow relationship between users is regarded as an edge E in the social network graph. That is, if two users follow each other, the two corresponding nodes in the social network graph will construct an edge (v i ,v j )∈E.

[0099] Step 1.2: Calculate user sentiment polarity;

[0100] The posts and comments of users in the social network are fed into the Transformer model to obtain the vector representation of the user's post text. Then, the final vector representation is passed through the Sigmoid layer to obtain the user v i The emotional polarity senti(v i ), senti(v i )∈[-1,1], when senti(vi )=-1, it means node user v i Extremely negative attitude towards public opinion, senti(v i )=1, it means node user v i Have a very positive attitude towards public opinion. i )=0, it means that the node user v i Maintain a neutral attitude towards public opinion or make no statement.

[0101] Step 1.3: Calculate user influence;

[0102] We use PageRank and sentiment polarity to quantify the influence of users in public opinion events. i ) is calculated as follows:

[0103] Influence(v i )=|senti(v i )*σ(PageRank(v i ))|;

[0104] Among them, senti(v i ) represents the user's sentiment polarity obtained in step 1.1, PageRank(v i ) represents the PageRank value of the user, σ(·) represents the sigmoid function, and the absolute value is added to ensure that the influence is a positive number.

[0105] Step 1.4: Community division;

[0106] The network is divided into multiple communities using the Girvan Newman algorithm, which includes the following steps: (1) calculating the betweenness of all existing edges in the network; (2) removing the edge with the highest betweenness; (3) recalculating the betweenness of all edges; and (4) repeating steps 2 and 3 until there are no edges left.

[0107] Step 1.5: Discovery of opinion leaders in the community;

[0108] In the communities divided in step 1.4, the influence of all nodes is sorted according to the user influence calculated in step 1.3, and the node with the greatest influence in each community is found and regarded as the opinion leader of the community. Then, according to the emotional polarity of each opinion leader, they are divided into active guides and passive recipients.

[0109] Step 2: Guide public opinion through the dual model;

[0110] The purpose of discovering opinion leaders in the community is to find the targets of implementation (active leaders) and the targets of being guided (passive recipients). Design a public opinion guidance algorithm based on dual learning.

[0111] Public opinion guidance specifically includes the following seven sub-steps:

[0112] Step 2.1: Construct a learning model and a dual model. The learning model is responsible for calculating the guidance weights of the active guide for different passive recipients, and the dual model is responsible for calculating the guided effect of the passive recipients after guidance as feedback to guide the next step of guidance.

[0113] Step 2.2: Input the network state into the learning model to obtain the guidance weights for the passive receivers.

[0114] Step 2.3: Conduct public opinion guidance for passive recipients based on guidance weights.

[0115] Step 2.4: Update the opinion of each node in the network according to the opinion evolution model.

[0116] Step 2.5: Based on each updated viewpoint, obtain the new network state according to the dual model.

[0117] Step 2.6: Return to step 4.2 until the opinions of all nodes in the network converge, at which point the bootstrapping process is complete.

[0118] This example uses positive node rate and average opinion value to evaluate the performance of the proposed method.

[0119] (1) The calculation of positive node rate is as follows:

[0120]

[0121] Where P pos Represents the positive node rate, which is used to measure the quality of the guidance results. pos Indicates the number of positive nodes at the end of the boot, N total Represents the number of nodes in the entire network. The larger the value of the positive node rate, the better the result of public opinion guidance. The specific results are shown in Table 2.

[0122] Table 2 Experimental results on Weibo-Flickr network data

[0123]

[0124] Experimental results show that the proposed method has significant advantages over the classical algorithm.

[0125] It is understandable that:

[0126] The present invention discloses a public opinion guidance method based on dual learning. The method combines the user's text information, can effectively capture the emotional polarity of users in social networks, and accurately quantify the user's influence in public opinion events.

[0127] The present invention uses dual learning to guide public opinion. Compared with traditional static methods, it can effectively handle changes in the network and make better decisions.

[0128] The present invention uses graph embedding technology to solve the problem that traditional methods cannot be migrated between different networks. Our method can migrate strategies on small-scale networks to large-scale networks and has high scalability.

[0129] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0130] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for positively guiding public opinion based on dual learning, characterized in that: The method comprises: Construct a social network graph and preprocess it to obtain community division results and user sentiment polarity; Using PageRank algorithm and combining users’ sentiment polarity, we search for opinion leaders in the community among the community division results. Establish a dual public opinion positive guidance model to guide the public opinion of opinion leaders in the community; use opinion dynamics and the dual public opinion positive guidance model to guide the nodes around the negative receivers, so that the state of the entire social network graph becomes positive, and judge the state of the social network graph according to the set proportion of positive state nodes, so as to achieve positive guidance of public opinion; After searching and obtaining the opinion leaders of the community in the community division results, the method further includes: distinguishing the opinion leaders of the community in the community division results according to the emotion polarity of the users, and obtaining the active guides and passive receivers in the opinion leader area of ​​the community; According to the dual learning method in machine learning, a dual public opinion positive guidance model is constructed: first, a dual task is set, and a guidance weight is assigned to the active guide to guide the passive receiver; the emotional polarity state change information of the passive receiver and the state change information of the network are fed back to the active guide; given the characteristic parameters of the dual public opinion positive guidance model, public opinion positive guidance is performed, and if the emotional polarity state change information of the passive receiver and the state change information of the network do not become positive, this process is repeated until the emotional polarity state change information of the passive receiver and the state change information of the network become positive, and the public opinion positive guidance is successful.

2. According to claim 1, a method for positively guiding public opinion based on dual learning is characterized in that: Crawl social network data, analyze social network data through the preset networkX library, and build a social network graph.

3. According to the method of public opinion positive guidance based on dual learning according to claim 1, it is characterized in that: The Girvan Newman algorithm is used to divide the social network graph into communities and obtain the community division results; among them, the community division uses modularity as the update indicator.

4. According to claim 1, a method for positively guiding public opinion based on dual learning is characterized in that: The user features in the social network graph are used to input the user features into a preset feedforward neural network to obtain the user's emotional polarity.

5. According to claim 1, a method for positively guiding public opinion based on dual learning is characterized in that: The opinion leaders of the community in the community division results obtained by the search specifically include: Use the PageRank algorithm in the preset networkX library to calculate the nodes in the social network graph and obtain the PageRank value of each node in the social network graph; The PageRank value of each node in the social network graph is combined with the above-obtained user sentiment polarity to obtain the user influence score; The user influence scores are sorted to obtain the opinion leaders of the community in the community division results.

6. A public opinion positive guidance system based on dual learning, characterized in that: The system comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors execute a method for positively guiding public opinion based on dual learning as described in any one of claims 1 to 5.

Images