A social network public opinion propagation key user mining method and system

Abstract

The application discloses a social network public opinion propagation key user mining method and system, relates to the technical field of network public opinion analysis, and comprises the following steps: obtaining historical social network public opinion propagation data and a preset propagation topology graph; according to the historical social network public opinion propagation data, performing key user mining on the preset propagation topology graph to obtain a candidate key user set; calculating a comment area atmosphere twist score and a twisted propagation benefit of each candidate key user in the candidate key user set; and according to the comment area atmosphere twist score and the twisted propagation benefit of each candidate key user, respectively judging whether each candidate key user is a key user of the preset propagation topology graph. The application can mine out a key user group which has the function of twisting a negative public opinion atmosphere and has a good inhibitory effect on further propagation of negative public opinion.

Description

Technical Field

[0001] This invention relates to the field of social network public opinion dissemination technology, and in particular to a method and system for mining key users in social network public opinion dissemination. Background Technology

[0002] Social network public opinion dissemination refers to the process by which public attitudes and opinions regarding specific events flow and spread within social networks. It is characterized by its susceptibility to the influence of opinion leaders and the vulnerability of its dissemination path to disruption. In real-world networks, public opinion-related information often ferments and evolves through interaction, even triggering group resonance or controversy, potentially impacting social cognition and decision-making. Therefore, complex network models, such as degree centrality, betweenness centrality, proximity centrality, and independent cascade models, are commonly used to identify key users in public opinion dissemination, providing foundational information for further intervention.

[0003] Mainstream methods for identifying key users in public opinion dissemination based on complex network models and pre-defined propagation topologies generally suffer from two common problems. First, network propagation is not static; once intervention occurs, a new propagation topology may form, thus diminishing the reliability of the algorithm model applied to the original propagation network. In fact, algorithms based on network topology generally have poor stability, determined by their inherent logic, making them less suitable for current application scenarios. Second, intervention should not necessarily target the most critical set of nodes themselves, as these users and their fan groups may have fickle or unwavering opinions. For the latter, in practice, significant effort may be invested in influencing the public opinion atmosphere under their posts, but the target account itself or its fan group may not agree, thus failing to achieve the desired effect of influencing the public opinion atmosphere and the development of public opinion dissemination. Summary of the Invention

[0004] To address the technical problems existing in the background art, this invention proposes a method and system for mining key users in social network public opinion dissemination.

[0005] Firstly, the present invention proposes a method for mining key users in social network public opinion dissemination, which includes:

[0006] Acquire historical social network public opinion dissemination data and preset dissemination topology maps;

[0007] Based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology to obtain a set of candidate key users;

[0008] Calculate the comment section atmosphere reversal score and post-reversal propagation benefit for each candidate key user in the candidate key user set;

[0009] Based on the comment section atmosphere reversal score and the propagation benefit after reversal of each candidate key user, determine whether each candidate key user is a key user in the preset propagation topology.

[0010] Preferably, based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology to obtain a set of candidate key users, specifically including:

[0011] Based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology using various algorithms in the preset network algorithm set to obtain the key user set for each algorithm.

[0012] The union of the key user sets of each algorithm is taken as the candidate key user set.

[0013] Preferably, the comment section atmosphere reversal score for each candidate key user in the candidate key user set is calculated, specifically including:

[0014] Step a: For each candidate key user in the candidate key user set, extract several original posts and comments of that user from historical social network public opinion dissemination data, and construct a comment sequence;

[0015] Step b: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the comment sequence, and record the percentage of positive labels and the percentage of negative labels.

[0016] Step c: Call the preset large language model to generate prompt words based on preset positive comments to simulate several virtual positive comments for several original posts of the user, and randomly insert several virtual positive comments into the comment sequence to obtain an extended comment sequence;

[0017] Step d: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the extended comment sequence, and record the percentage of positive labels and the percentage of negative labels;

[0018] Step e: Determine if the preset number of virtual positive comments generated is met; if not, proceed to step c; if yes, proceed to step f.

[0019] Step f: Calculate the simulated gains of the user in each virtual positive comment generation process, and select the maximum value of the simulated gains of the user in all virtual positive comment generation processes as the user's comment section atmosphere reversal score.

[0020] Preferably, each candidate key user The simulated benefit during the i-th virtual positive review generation process is:

[0021] ;

[0022] In the formula, The simulated benefit for the current user during the i-th virtual positive review generation process is μ, where μ is a preset coefficient; i = 1, 2, ..., N, and N is the preset number of virtual positive review generation times. The percentage of positive labels. The percentage of negative labels; Let represent the number of virtual positive comments generated during the i-th virtual positive comment generation process.

[0023] Preferably, the calculation process for the post-propagation benefit of each candidate key user in the candidate key user set includes:

[0024] The preset network propagation topology is divided into communities, and the preset propagation model is run on the preset network propagation topology after community division to obtain the first propagation result;

[0025] Each candidate key user is removed from the preset network propagation topology to obtain multiple intermediate network propagation topology diagrams.

[0026] Each intermediate network propagation topology is divided into communities, and the preset propagation model is run several times on each intermediate network propagation topology after community division to obtain the second propagation result of each intermediate network propagation topology.

[0027] Based on the first propagation result and the second propagation result of each intermediate network propagation topology, calculate the profit of each intermediate network propagation topology.

[0028] Based on the revenue of each intermediate network propagation topology, calculate the reverse propagation revenue of each excluded candidate key user corresponding to each intermediate network propagation topology.

[0029] Preferably, the benefits of each intermediate network propagation topology include: a reduction in negative user reach, an increase in the time of negative peak arrival, and a reduction in the amount of negative information received across communities;

[0030] The post-propagation gain for each candidate key user is the weighted sum of the gains of each item in the corresponding intermediate network propagation topology.

[0031] Preferably, ;

[0032] ;

[0033] ;

[0034] ;

[0035] In the formula, The value that reduces the reach of negative users; The degree of increase in the arrival time of the negative peak; To reduce the amount of negative information received across social media platforms; , and These represent, in turn, the negative user reach rate, the peak arrival time of negative information, and the amount of negative information from community j to community i in the first dissemination result; For the community , For the community , To pre-define the network propagation topology, To preset the number of nodes in the network propagation topology, Let be the number of nodes in community j. For the community The number of nodes, The community set is obtained by dividing a pre-defined network propagation topology into communities; , and These represent, in turn, the negative user reach rate, the global negative peak arrival time, and the amount of negative information from community j in group i; The post-spread benefits for current key candidate users; and These are preset weights.

[0036] Preferably, based on the comment section atmosphere reversal score and the propagation benefit after reversal of each candidate key user, it is determined whether each candidate key user is a key user in the preset propagation topology, specifically including:

[0037] Determine whether the comment section atmosphere reversal score and the propagation benefit after reversal are uniformly positive for each candidate key user; if not, determine that the user is not a key user in the preset propagation topology; if so, calculate the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user.

[0038] Determine whether the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user is greater than the preset total benefit threshold; if so, determine that the user is a key user in the preset propagation topology; if not, determine that the user is not a key user in the preset propagation topology.

[0039] Secondly, this invention also proposes a key user mining system for social network public opinion dissemination, comprising:

[0040] The acquisition module is used to acquire historical social network public opinion dissemination data and preset dissemination topology maps;

[0041] The key user mining module is used to mine key users from a preset propagation topology based on historical social network public opinion propagation data, and obtain a set of candidate key users.

[0042] The processing module is used to calculate the comment area atmosphere reversal score and the propagation benefit after reversal for each candidate key user in the candidate key user set; based on the comment area atmosphere reversal score and the propagation benefit after reversal for each candidate key user, it determines whether each candidate key user is a key user in the preset propagation topology.

[0043] Preferably, based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology to obtain a set of candidate key users, specifically including:

[0044] Based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology using various algorithms in the preset network algorithm set to obtain the key user set for each algorithm.

[0045] The union of the key user sets of each algorithm is taken as the candidate key user set.

[0046] Preferably, the comment section atmosphere reversal score for each candidate key user in the candidate key user set is calculated, specifically including:

[0047] Step a: For each candidate key user in the candidate key user set, extract several original posts and comments of that user from historical social network public opinion dissemination data, and construct a comment sequence;

[0048] Step b: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the comment sequence, and record the percentage of positive labels and the percentage of negative labels.

[0049] Step c: Call the preset large language model to generate prompt words based on preset positive comments to simulate several virtual positive comments for several original posts of the user, and randomly insert several virtual positive comments into the comment sequence to obtain an extended comment sequence;

[0050] Step d: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the extended comment sequence, and record the percentage of positive labels and the percentage of negative labels;

[0051] Step e: Determine if the preset number of virtual positive comments generated is met; if not, proceed to step c; if yes, proceed to step f.

[0052] Step f: Calculate the simulated gains of the user in each virtual positive comment generation process, and select the maximum value of the simulated gains of the user in all virtual positive comment generation processes as the user's comment section atmosphere reversal score.

[0053] Preferably, the calculation process for the post-propagation benefit of each candidate key user in the candidate key user set includes:

[0054] The preset network propagation topology is divided into communities, and the preset propagation model is run on the preset network propagation topology after community division to obtain the first propagation result;

[0055] Each candidate key user is removed from the preset network propagation topology to obtain multiple intermediate network propagation topology diagrams.

[0056] Each intermediate network propagation topology is divided into communities, and the preset propagation model is run several times on each intermediate network propagation topology after community division to obtain the second propagation result of each intermediate network propagation topology.

[0057] Based on the first propagation result and the second propagation result of each intermediate network propagation topology, calculate the profit of each intermediate network propagation topology.

[0058] Based on the revenue of each intermediate network propagation topology, calculate the reverse propagation revenue of each excluded candidate key user corresponding to each intermediate network propagation topology.

[0059] Preferably, the benefits of each intermediate network propagation topology include: a reduction in negative user reach, an increase in the time of negative peak arrival, and a reduction in the amount of negative information received across communities;

[0060] The post-propagation gain for each candidate key user is the weighted sum of the gains of each item in the corresponding intermediate network propagation topology.

[0061] Preferably, based on the comment section atmosphere reversal score and the propagation benefit after reversal of each candidate key user, it is determined whether each candidate key user is a key user in the preset propagation topology, specifically including:

[0062] Determine whether the comment section atmosphere reversal score and the propagation benefit after reversal are uniformly positive for each candidate key user; if not, determine that the user is not a key user in the preset propagation topology; if so, calculate the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user.

[0063] Determine whether the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user is greater than the preset total benefit threshold; if so, determine that the user is a key user in the preset propagation topology; if not, determine that the user is not a key user in the preset propagation topology.

[0064] The proposed method and system for mining key users in social network public opinion dissemination in this invention first mines key users from a preset dissemination topology based on historical social network public opinion dissemination data, obtaining a set of candidate key users; then, it calculates the comment area atmosphere reversal score and the dissemination benefit after reversal for each candidate key user in the set; based on the comment area atmosphere reversal score and the dissemination benefit after reversal for each candidate key user, it determines whether each candidate key user is a key user in the preset dissemination topology. This method can identify key user groups that have a good effect on reversing negative public opinion atmosphere and inhibiting the further spread of negative public opinion, which is of great help in making contingency plans and responses in advance, guiding the development of public opinion trends, and maintaining the online public opinion environment. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the structure of a method and system for mining key users in social network public opinion dissemination proposed in this invention. Detailed Implementation

[0066] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0067] Reference Figure 1 This invention proposes a method and system for mining key users in social network public opinion dissemination, comprising:

[0068] Acquire historical social network public opinion dissemination data and preset dissemination topology maps;

[0069] Based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology to obtain a set of candidate key users;

[0070] Calculate the comment section atmosphere reversal score for each candidate key user in the candidate key user set;

[0071] Calculate the turnaround propagation benefit for each candidate key user in the candidate key user set;

[0072] Based on the comment section atmosphere reversal score and the propagation benefit after reversal of each candidate key user, determine whether each candidate key user is a key user in the preset propagation topology.

[0073] In the process of mining the spread of online public opinion events, this invention first mines key users based on historical social network public opinion spread data, obtaining a set of candidate key users; then, it calculates the comment area atmosphere reversal score and the spread benefit after reversal for each candidate key user in the set; based on the comment area atmosphere reversal score and the spread benefit after reversal for each candidate key user, it determines whether each candidate key user is a key user in the preset spread topology. This invention can identify key user groups that have a good effect on reversing negative public opinion atmosphere and inhibiting the further spread of negative public opinion, which is of great help in making contingency plans and responses in advance, guiding the development of public opinion trends, and maintaining the online public opinion environment.

[0074] To obtain a comprehensive key user mining framework, this embodiment performs key user mining on a preset propagation topology based on historical social network sentiment dissemination data, resulting in a candidate key user set, specifically including:

[0075] Based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology using various algorithms in the preset network algorithm set to obtain the key user set for each algorithm.

[0076] The union of the key user sets of each algorithm is taken as the candidate key user set.

[0077] Specifically, a preset set of network algorithms is defined as A = {degree centrality, betweenness centrality, proximity centrality, independent concatenation, ...}, which can be flexibly selected as needed.

[0078] For the node set is The preset propagation topology has the following set of candidate key users:

[0079] ;

[0080] In the formula, This represents the set of candidate key users, which is the union of the sets of k-key users obtained from different preset network algorithms. For algorithm a, k-key user combination, Let be the scoring function for algorithm a, such as degree, PageRank value, or expected propagation range.

[0081] To provide more candidates for subsequent mining of key user sets, k can be appropriately increased in this step. Subsequent steps will... The analysis is performed within the scope of the set.

[0082] It's important to understand that key users in the development and spread of public opinion are often "big V" accounts with a large following. When they post information related to a particular public opinion event, numerous comments from their followers will follow. In this context, the most common external intervention method for a negative public opinion event is to proactively post more positive comments. However, the atmosphere of the comment section—that is, whether the opinions, stances, and emotions surrounding the public opinion can be reversed—is not considered in traditional key node mining based on complex network models.

[0083] To address this issue, this embodiment calculates the comment section atmosphere reversal score for each candidate key user in the candidate key user set, specifically including:

[0084] Step a: For each candidate key user in the candidate key user set, extract several original posts and comments of that user from historical social network public opinion dissemination data, and construct a comment sequence;

[0085] Step b: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the comment sequence, and record the percentage of positive labels and the percentage of negative labels.

[0086] Step c: Call the preset large language model to generate prompt words based on preset positive comments to simulate several virtual positive comments for several original posts of the user, and randomly insert several virtual positive comments into the comment sequence to obtain an extended comment sequence;

[0087] Step d: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the extended comment sequence, and record the percentage of positive labels and the percentage of negative labels;

[0088] Step e: Determine if the preset number of virtual positive comments generated is met; if not, proceed to step c; if yes, proceed to step f.

[0089] Step f: Calculate the simulated gains of the user in each virtual positive comment generation process, and select the maximum value of the simulated gains of the user in all virtual positive comment generation processes as the user's comment section atmosphere reversal score.

[0090] Among them, each candidate key user The simulated benefit during the i-th virtual positive review generation process is: ;

[0091] In the formula, This represents the simulated benefit for the current user during the i-th virtual positive review generation process. The value is a preset coefficient, determined according to the actual situation, to reflect the cost of posting; i=1,2,…,N, where N is the preset number of virtual positive comments generated; The percentage of positive labels. The percentage of negative labels; Let represent the number of virtual positive comments generated during the i-th virtual positive comment generation process.

[0092] It is important to note that Maximum value in sequence It can be a positive value or a negative value, i.e., a negative return.

[0093] It's important to note that generating positive comments requires understanding the public opinion context and reading all comments, while also meeting constraints such as mimicking fan styles and similar comment length. Different batches of virtual positive comments are generated independently, and virtual positive comments can be selected repeatedly.

[0094] It should be understood that public opinion intervention is primarily aimed at mitigating the impact of negative voices. With gentle intervention, efforts can be made to shift the atmosphere of influential figures' comment sections to a more positive one, thereby influencing the spread of information.

[0095] Therefore, this embodiment needs to ignore the promoting effect of positive public opinion and focus on the spread of negative public opinion. Let's assume a user... The atmosphere in the comment section was completely reversed to a positive one. This user can be considered to have been removed from the propagation graph, which is equivalent to removing the user from the preset propagation topology graph G. and related edges.

[0096] To quantitatively evaluate and eliminate candidate key users The extent to which negative public opinion is weakened globally provides a basis for subsequent public opinion intervention. This embodiment will exclude nodes. The impact can be broken down into the following dimensions: First, the proportion of negative information ultimately reaching users decreases, denoted as... Secondly, the arrival time of the negative peak is delayed. In practice, this is often accompanied by a decrease in the intensity of public opinion, denoted as... .

[0097] Meanwhile, considering the significant randomness inherent in examining the simulation at the individual node level, an examination at the community level is conducted to improve the stability and reliability of the simulation. Additionally, a cross-community penetration flux indicator is added, denoted as... This is used to reflect the decrease in the amount of negative information flowing into this community from other communities, thus reflecting the current level of interference in the community.

[0098] Specifically, firstly, a pre-defined community partitioning model is used to partition the pre-defined network propagation topology graph G into communities, resulting in a community set that includes several communities. ;

[0099] Secondly, a preset propagation model, such as the mainstream model of independent cascading, is run several times on the preset network propagation topology G to obtain the first propagation result; the first propagation result includes the global negative user reach rate. The global negative peak arrival time is The amount of negative information from community j to community i is ;in, For the community , For the community ;

[0100] Next, each candidate key user is excluded from the preset network propagation topology graph G to obtain the intermediate network propagation topology graph corresponding to each candidate key user. ;in, In the formula, This indicates excluded candidate key users and related edges;

[0101] The propagation topology of each intermediate network is analyzed using a pre-defined community segmentation model. The community is divided, and the topology diagram is propagated in each intermediate network after the community is divided. Running the same pre-set propagation model several times yields a second propagation result; this second propagation result includes the global negative user reach rate. The global negative peak arrival time is The amount of negative information from community j to community i is ;

[0102] Calculate the propagation topology of each intermediate network separately. Benefits across various dimensions; where each intermediate network propagation topology is represented. Benefits across various dimensions include a reduction in the rate of negative user reach. The degree of increase in the arrival time of negative peaks And the degree of reduction of negative information across communities ;

[0103] in, ; ; ;

[0104] in, Calculate the number of nodes in a graph or community; This reflects a decrease in the rate of reaching negative users; This reflects the increased degree of delay in the arrival time of the negative peak; This reflects the degree of reduction in receiving negative information across social media platforms.

[0105] The propagation gain after reversal is calculated for the intermediate network propagation topology after excluding each candidate key user; specifically, the propagation gain after reversal is as follows:

[0106] ;

[0107] in, and These are the preset weights for each dimension. To reverse the spread of benefits.

[0108] It should be noted that the propagation benefit after reversal may be positive or negative, i.e., a negative benefit.

[0109] The preset community segmentation model in this embodiment adopts mainstream community segmentation models such as Louvian.

[0110] In other words, the calculation process for the post-propagation gain of each candidate key user in the candidate key user set in this embodiment includes:

[0111] The preset network propagation topology is divided into communities, and the preset propagation model is run on the preset network propagation topology after community division to obtain the first propagation result;

[0112] Each candidate key user is removed from the preset network propagation topology to obtain multiple intermediate network propagation topology diagrams.

[0113] Each intermediate network propagation topology is divided into communities, and the preset propagation model is run several times on each intermediate network propagation topology after community division to obtain the second propagation result of each intermediate network propagation topology.

[0114] Based on the first propagation result and the second propagation result of each intermediate network propagation topology, calculate the profit of each intermediate network propagation topology.

[0115] Based on the revenue of each intermediate network propagation topology, calculate the reverse propagation revenue of each excluded candidate key user corresponding to each intermediate network propagation topology.

[0116] The benefits of each intermediate network propagation topology include: the reduction in negative user reach, the increase in negative peak arrival time, and the reduction in the reception of negative information across communities.

[0117] The post-propagation gain for each candidate key user is the weighted sum of the gains of each item in the corresponding intermediate network propagation topology.

[0118] Run the preset propagation model once or multiple times. When running it multiple times, take the average value.

[0119] In this embodiment, based on the comment section atmosphere reversal score and the propagation benefit after reversal of each candidate key user, it is determined whether each candidate key user is a key user in the preset propagation topology, specifically including:

[0120] Determine whether the comment section atmosphere reversal score and the propagation benefit after reversal are uniformly positive for each candidate key user; if not, determine that the user is not a key user in the preset propagation topology; if so, calculate the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user.

[0121] Determine whether the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user is greater than the preset total benefit threshold; if so, determine that the user is a key user in the preset propagation topology; if not, determine that the user is not a key user in the preset propagation topology.

[0122] In other words, key users in the pre-defined propagation topology must simultaneously meet the following conditions:

[0123] ;

[0124] That is, the individual benefit in each dimension is positive, and the sum of the comment section atmosphere reversal score and the propagation benefit after the reversal for each candidate key user, i.e., the total benefit, meets the preset total benefit threshold. .

[0125] Secondly, this invention also proposes a key user mining system for social network public opinion dissemination, comprising:

[0126] The acquisition module is used to acquire historical social network public opinion dissemination data and preset dissemination topology maps;

[0127] The key user mining module is used to mine key users from a preset propagation topology based on historical social network public opinion propagation data, and obtain a set of candidate key users.

[0128] The processing module is used to calculate the comment area atmosphere reversal score and the propagation benefit after reversal for each candidate key user in the candidate key user set; based on the comment area atmosphere reversal score and the propagation benefit after reversal for each candidate key user, it determines whether each candidate key user is a key user in the preset propagation topology.

[0129] In this embodiment, based on historical social network public opinion propagation data, key user mining is performed on a preset propagation topology to obtain a candidate key user set, specifically including:

[0130] Based on historical social network public opinion dissemination data, key user mining is performed on the preset dissemination topology using various algorithms in the preset network algorithm set to obtain the key user set for each algorithm.

[0131] The union of the key user sets of each algorithm is taken as the candidate key user set.

[0132] In this embodiment, the comment section atmosphere reversal score for each candidate key user in the candidate key user set is calculated, specifically including:

[0133] Step a: For each candidate key user in the candidate key user set, extract several original posts and comments of that user from historical social network public opinion dissemination data, and construct a comment sequence;

[0134] Step b: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the comment sequence, and record the percentage of positive labels and the percentage of negative labels.

[0135] Step c: Call the preset large language model to generate prompt words based on preset positive comments to simulate several virtual positive comments for several original posts of the user, and randomly insert several virtual positive comments into the comment sequence to obtain an extended comment sequence;

[0136] Step d: Use a pre-defined large language model to assign a positive, neutral, or negative label to each comment in the extended comment sequence, and record the percentage of positive labels and the percentage of negative labels;

[0137] Step e: Determine if the preset number of virtual positive comments generated is met; if not, proceed to step c; if yes, proceed to step f.

[0138] Step f: Calculate the simulated gains of the user in each virtual positive comment generation process, and select the maximum value of the simulated gains of the user in all virtual positive comment generation processes as the user's comment section atmosphere reversal score.

[0139] In this embodiment, the calculation process for the post-propagation gain of each candidate key user in the candidate key user set includes:

[0140] The preset network propagation topology is divided into communities, and the preset propagation model is run on the preset network propagation topology after community division to obtain the first propagation result;

[0141] Each candidate key user is removed from the preset network propagation topology to obtain multiple intermediate network propagation topology diagrams.

[0142] Each intermediate network propagation topology is divided into communities, and the preset propagation model is run several times on each intermediate network propagation topology after community division to obtain the second propagation result of each intermediate network propagation topology.

[0143] Based on the first propagation result and the second propagation result of each intermediate network propagation topology, calculate the profit of each intermediate network propagation topology.

[0144] Based on the revenue of each intermediate network propagation topology, calculate the reverse propagation revenue of each excluded candidate key user corresponding to each intermediate network propagation topology.

[0145] Preferably, the benefits of each intermediate network propagation topology include: a reduction in negative user reach, an increase in the time of negative peak arrival, and a reduction in the amount of negative information received across communities;

[0146] The post-propagation gain for each candidate key user is the weighted sum of the gains of each item in the corresponding intermediate network propagation topology.

[0147] In this embodiment, based on the comment section atmosphere reversal score and the propagation benefit after reversal of each candidate key user, it is determined whether each candidate key user is a key user in the preset propagation topology, specifically including:

[0148] Determine whether the comment section atmosphere reversal score and the propagation benefit after reversal are uniformly positive for each candidate key user; if not, determine that the user is not a key user in the preset propagation topology; if so, calculate the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user.

[0149] Determine whether the sum of the comment section atmosphere reversal score and the propagation benefit after reversal for each candidate key user is greater than the preset total benefit threshold; if so, determine that the user is a key user in the preset propagation topology; if not, determine that the user is not a key user in the preset propagation topology.

[0150] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for mining key users of social network public opinion propagation, characterized in that, The method comprises the following steps: obtaining historical social network public opinion propagation data and a preset propagation topology graph; mining key users from the preset propagation topology graph according to the historical social network public opinion propagation data to obtain a candidate key user set; calculating a comment area atmosphere twist score and a twisted propagation benefit of each candidate key user in the candidate key user set; determining whether each candidate key user is a key user of the preset propagation topology graph according to the comment area atmosphere twist score and the twisted propagation benefit of each candidate key user; wherein the calculation process of the comment area atmosphere twist score comprises: step a, for each candidate key user, extracting a plurality of original posts and comments of the user from the historical social network public opinion propagation data, and constructing a comment sequence; step b, using a preset large language model to label each comment in the comment sequence as positive, neutral or negative, and recording the proportion of positive labels and the proportion of negative labels; step c, calling the preset large language model to simulate generating a plurality of virtual positive comments for the plurality of original posts of the user according to a preset positive comment generation prompt word, and randomly inserting the plurality of virtual positive comments into the comment sequence to obtain an extended comment sequence; step d, using the preset large language model to label each comment in the extended comment sequence as positive, neutral or negative, and recording the proportion of positive labels and the proportion of negative labels; step e, determining whether the preset virtual positive comment generation frequency is met; if not, proceeding to step c; if yes, proceeding to step f; step f, calculating the simulated benefit of the user in each virtual positive comment generation process, and selecting the maximum value of the simulated benefit as the comment area atmosphere twist score of the user; wherein the calculation process of the twisted propagation benefit comprises: dividing the preset network propagation topology graph into communities and running a preset propagation model thereon to obtain a first propagation result; removing each candidate key user from the preset network propagation topology graph to obtain a plurality of intermediate network propagation topology graphs; dividing each intermediate network propagation topology graph into communities, and running the preset propagation model thereon for a plurality of times to obtain a second propagation result of each intermediate network propagation topology graph; calculating the benefit of each intermediate network propagation topology graph and the twisted propagation benefit of each candidate key user corresponding to each intermediate network propagation topology graph according to the first propagation result and the second propagation result of each intermediate network propagation topology graph. 2.The social network public opinion propagation key user mining method according to claim 1, characterized in that, mining key users from the preset propagation topology graph according to the historical social network public opinion propagation data to obtain a candidate key user set, specifically comprising: mining key users from the preset propagation topology graph using each algorithm in a preset set of network algorithms according to the historical social network public opinion propagation data to obtain a key user set of each algorithm; taking the union of the key user sets of each algorithm as the candidate key user set.

3. The social network opinion propagation key user mining method of claim 1, wherein, the simulated benefit of each candidate key user in each virtual positive comment generation process is: ； In the formula, is the simulation income of the current user in the i-th virtual positive review generation process, μ is a preset coefficient; i=1, 2, …, N, and N is a preset virtual positive review generation number; is the proportion of positive labels, is the proportion of negative labels. is the number of virtual positive reviews generated in the i-th virtual positive review generation process.

4. The social network opinion propagation key user mining method of claim 1, wherein, the benefit of each intermediate network propagation topology graph includes: a negative user reach rate reduction value, an increase degree of negative peak arrival time, and a reduction degree of accepting cross-community negative information; The twist post-propagation benefit of each candidate key user is a weighted sum of respective benefits of the corresponding intermediate network propagation topology graph.

5. The social network public opinion propagation key user mining method of claim 4, wherein, ； In the formula, The value that reduces the reach of negative users; The degree of increase in the arrival time of the negative peak; To reduce the amount of negative information received across social media platforms; , and These represent, in turn, the negative user reach rate, the peak arrival time of negative information, and the amount of negative information from community j to community i in the first dissemination result; For community i, For community j, To pre-define the network propagation topology, To preset the number of nodes in the network propagation topology, Let be the number of nodes in community j. Let be the number of nodes in community i. The community set is obtained by dividing a pre-defined network propagation topology into communities; , and These represent, in turn, the negative user reach rate, the global negative peak arrival time, and the amount of negative information from community j to community i; The post-spread benefits for current key candidate users; and These are preset weights.

6. The social network opinion propagation key user mining method of claim 4, wherein, According to the comment area atmosphere twist score and the twist post-propagation benefit of each candidate key user, it is respectively judged whether each candidate key user is a key user of the preset propagation topology graph, specifically including: It is judged whether the comment area atmosphere twist score and the twist post-propagation benefit of each candidate key user are uniformly positive; if not, it is determined that the user is not a key user of the preset propagation topology graph; if yes, the sum of the comment area atmosphere twist score and the twist post-propagation benefit of each candidate key user is calculated; It is judged whether the sum of the comment area atmosphere twist score and the twist post-propagation benefit of each candidate key user is greater than a preset total benefit threshold; if yes, it is determined that the user is a key user of the preset propagation topology graph; if not, it is determined that the user is not a key user of the preset propagation topology graph. 7.A system for mining key users of social network opinion propagation, characterized in that, including: An acquisition module is configured to acquire historical social network public opinion propagation data and a preset propagation topology graph; A key user mining module is configured to mine key users of the preset propagation topology graph according to the historical social network public opinion propagation data, to obtain a candidate key user set; A processing module is configured to calculate a comment area atmosphere twist score and a twist post-propagation benefit of each candidate key user in the candidate key user set; and to judge whether each candidate key user is a key user of the preset propagation topology graph according to the comment area atmosphere twist score and the twist post-propagation benefit of each candidate key user. The calculation process of the comment area atmosphere twist score includes: Step a, for each candidate key user, a plurality of original posts and comments of the user are extracted from the historical social network public opinion propagation data, and a comment sequence is constructed; Step b, a preset large language model is used to label each comment in the comment sequence as positive, neutral or negative, and the proportion of positive labels and the proportion of negative labels are recorded; Step c, the preset large language model is called to simulate a plurality of virtual positive comments for the plurality of original posts of the user according to a preset positive comment generation prompt word, and the plurality of virtual positive comments are randomly inserted into the comment sequence to obtain an extended comment sequence; Step d, the preset large language model is used to label each comment in the extended comment sequence as positive, neutral or negative, and the proportion of positive labels and the proportion of negative labels are recorded; Step e, it is judged whether the preset virtual positive comment generation number is met; if not, step c is entered; if yes, step f is entered; Step f, the simulation benefit of the user in each virtual positive comment generation process is calculated, and the maximum value of the simulation benefit is selected as the comment area atmosphere twist score of the user; The calculation process of the twist post-propagation benefit includes: The preset network propagation topology graph is divided into communities and a preset propagation model is run thereon to obtain a first propagation result; remove each candidate key user from the preset network propagation topology graph respectively to obtain a plurality of intermediate network propagation topology graphs; perform community division on each intermediate network propagation topology graph respectively, and run a preset propagation model on each intermediate network propagation topology graph for several times respectively to obtain a second propagation result of each intermediate network propagation topology graph; According to the first propagation result and the second propagation result of each intermediate network propagation topology graph, the revenue of each intermediate network propagation topology graph and the twisted propagation revenue of each candidate key user excluded by each intermediate network propagation topology graph are calculated respectively. 8.The social network public opinion propagation key user mining system of claim 7, wherein, According to the historical social network public opinion propagation data, the key user mining is performed on the preset propagation topology graph to obtain a candidate key user set, specifically including: According to the historical social network public opinion propagation data, each algorithm in the preset network algorithm set is used to perform key user mining on the preset propagation topology graph to obtain a key user set of each algorithm; Take the union of the key user sets of each algorithm as the candidate key user set.

Images