Network violent event responsibility quantification method and device

Abstract

The invention provides a network violence event responsibility quantification method and device. The method comprises the following steps: obtaining evidence obtaining data of a network violence event; extracting interaction behaviors of users participating in the network violence event according to the evidence obtaining data to construct a propagation path diagram, and expanding friend relationships of the users on the basis of a user set in the propagation path diagram to construct a social network sub-diagram focused on the network violence event; quantifying the social influence of each user in network violent event propagation according to the propagation path graph and the social network sub-graph; quantifying the direct responsibility of each user in the network violence event according to the scale of the network violence event and the speech aggressiveness, the participation intensity and the participation duration of the user; and synthesizing the social influence and the direct responsibility of the user to obtain the total responsibility of the user in the network violence event. By applying the method, panoramic and structured evidence fixation and data analysis can be carried out on network violence events, and infringement responsibilities can be scientifically and finely measured.

Description

Technical Field

[0001] This invention relates to the field of internet content security management technology, and in particular to a method and apparatus for quantifying responsibility for cyberbullying incidents. Background Technology

[0002] With the widespread use of internet social media platforms, cyberbullying incidents are frequent. These incidents are characterized by the decentralized nature of infringing statements (involving numerous perpetrators) and the collective nature of harm (individual statements cause minor damage, but collective statements cause enormous harm). Currently, victims face multiple difficulties in seeking redress and in regulatory agencies' handling of these cases.

[0003] For victims: the cost of protecting their rights is extremely high, and there are problems such as difficulty in obtaining evidence (dispersed opinions are easily deleted), difficulty in providing evidence (notarization is required and the process is cumbersome), and difficulty in quantifying (damage is difficult to measure using traditional property loss standards).

[0004] For platforms and regulatory agencies: Current technology makes it difficult to accurately distinguish the responsibilities of a massive number of participants. Punishments are often only based on a single dimension such as the severity of the content or the number of reposts, making it impossible to scientifically assess the actual influence and contribution of each infringer in the entire chain of events. This can easily lead to "the law not punishing the masses" or excessively lenient or harsh punishments, failing to achieve the governance goal of proportionality between offense and punishment.

[0005] Existing electronic evidence collection tools (such as the Power Chain) can only provide single-point and decentralized evidence collection functions. They cannot conduct a global and structured analysis of a complete cyberbullying incident, nor can they quantify and rank infringement liability. Their evidentiary value and ability to support refined management are limited.

[0006] Therefore, there is an urgent need for a method to quantify the responsibility for cyber violence incidents that can be automated, collect evidence in batches, and scientifically quantify the liability for infringement, in order to assist judicial decisions and platform governance. Summary of the Invention

[0007] The purpose of this invention is to provide a method and apparatus for quantifying liability in cyberbullying incidents, so as to scientifically quantify the tort liability in cyberbullying incidents.

[0008] Firstly, the method for quantifying responsibility for cyberbullying incidents provided by this invention includes: acquiring evidence data of cyberbullying incidents, including posting and interaction data of cyberbullying incidents, corresponding user information, and publishing metadata; based on the evidence data, extracting user interaction behaviors involved in cyberbullying incidents to construct a propagation path diagram, and expanding user friend relationships based on the user set in the propagation path diagram to construct a social network subgraph focusing on cyberbullying incidents; quantifying the social influence of each user in the propagation of cyberbullying incidents based on the propagation path diagram and the social network subgraph; quantifying the direct responsibility of each user in cyberbullying incidents based on the scale of cyberbullying incidents and the user's offensive language, participation intensity, and participation duration; and comprehensively considering the user's social influence and direct responsibility to obtain the user's total responsibility in cyberbullying incidents.

[0009] The beneficial effects of the method for quantifying liability in cyberbullying incidents provided by this invention are: it enables panoramic and structured evidence collection and data analysis of a cyberbullying incident, and scientifically and precisely measures liability for infringement. This measurement result integrates the degree of harm caused by the directly committed infringement and the indirect influence of the infringement on social networks, ultimately generating a highly valuable liability quantification report. This provides data-driven and visualized decision-making support for victim rights protection, platform handling, and judicial adjudication.

[0010] In one possible embodiment, obtaining evidence data for cyberbullying incidents includes: retrieving relevant social data of cyberbullying incidents from evidence data obtained using blockchain evidence storage technology according to set search conditions; the set search conditions are to use strong features of cyberbullying incidents for precise retrieval, and the strong features include the identity identifiers of the participants in the cyberbullying incident, topic identifiers, key account sets, and time ranges; when strong features are insufficient, key entities and time ranges are used as the main search criteria, and the information dissemination chain is expanded by combining social interaction behavior.

[0011] In another possible embodiment, user interaction behaviors involved in cyberbullying incidents are extracted to construct a propagation path graph. The user set in the propagation path graph is then used as a basis to expand user friend relationships to construct a social network subgraph focusing on cyberbullying incidents. This includes: using users involved in cyberbullying incidents from the evidence data as nodes, constructing directed edges from the originator to the interactor based on user interaction behaviors, and assigning weights to the edges according to the nature of the interaction to form a propagation path graph; using the user set in the propagation path graph as the base user set, extracting the users' friends from the social relationship graph of the social platform to form the users' friend set; and constructing a social network subgraph based on the base user set and the users' friend set. The extraction of users' friends from the social relationship graph of the social platform includes: extracting users who have a follow or being followed relationship with the user, or users whose historical interaction frequency with the user is greater than a set threshold.

[0012] In other possible embodiments, for a user, the set of users reachable from the user as the starting point in the propagation path graph is defined as the user's propagation influence set, and the union of the user's propagation influence set and the friend set forms the user's influenced user set; the social influence of each user in the propagation of cyberbullying events is quantified based on the propagation path graph and the social network subgraph, including: constructing a social feature vector based on the user's social attributes and historical indicators; quantifying the user's social characteristics based on the user's participation in cyberbullying events and their influence on users in the influenced user set; calculating the user's counterfactual behavior prediction difference using the user's social feature vector and social characteristics; weighting and summarizing the user's influence on each user node in the friend set based on the social network subgraph and the counterfactual behavior prediction difference to obtain the user's friend influence; setting a fixed basic influence weight; weighting and summarizing the user's influence on each non-friend user node in the influenced user set based on the user's participation in cyberbullying events and the basic influence weight to obtain the user's non-friend influence; and calculating the user's social influence by combining friend influence and non-friend influence.

[0013] The calculation of a user's social characteristics follows the formula: ,in, Indicates target user social characteristics, Indicates user For target users Friends Collection The nodes in Indicates user Participation in cyberbullying incidents Indicates user For users The edge weights in a social network subgraph; the calculation of the user's counterfactual behavior prediction difference satisfies the following formula: ,in, Indicates user The difference in counterfactual behavior prediction, Indicates user Social feature vectors, Indicates user In social characteristics Predicted participation Indicates user Counterfactual predictions of engagement when the influence of social context is removed.

[0014] The user's friend influence is calculated according to the following formula: ,in, Indicates user The influence of friends, Indicates user For users Friends Collection The nodes in Indicates user For users Edge weights in a social network subgraph Indicates user The difference in counterfactual behavior prediction, Indicates user The number of friends in a social network subgraph.

[0015] The user's influence beyond friends is calculated according to the following formula: ,in, Indicates user Non-friend influence, Indicates user For users The impact is on non-friend users in the user set. This indicates the set base influence weight. Indicates user Involvement in cyberbullying incidents.

[0016] The direct responsibility of each user in a cyberbullying incident is quantified based on the scale of the incident and the user's offensive language, participation intensity, and duration of participation. This includes: classifying the scale of a cyberbullying incident based on the number of participating users and the total amount of related content to obtain an incident scale coefficient; using a pre-trained semantic analysis model to identify the offensiveness of user-posted text content in the evidence data to obtain an offensiveness score, and calculating the user's offensive language score based on the offensiveness scores of all user-posted text content; determining the user's participation intensity based on the number of user-posted text content with offensiveness scores exceeding a set threshold; calculating the time span of user participation in the cyberbullying incident; and then normalizing the incident scale coefficient, offensive language score, participation intensity, and time span before performing a weighted sum to obtain a quantified value of the user's direct responsibility in the cyberbullying incident.

[0017] A user's total responsibility in a cyberbullying incident is the sum of their direct and indirect responsibilities. After obtaining the user's total responsibility in a cyberbullying incident, the process also includes: sorting the total responsibility quantification values ​​of all users who participated in the cyberbullying incident to generate a macro-level responsibility quantification ranking report; and generating an individual tort liability detail report based on the detailed information related to the direct and indirect responsibilities of each user who participated in the cyberbullying incident.

[0018] Secondly, the present invention also provides a device for quantifying responsibility for cyberbullying incidents, comprising: an evidence collection unit for acquiring evidence data of cyberbullying incidents, including posting and interaction data of cyberbullying incidents, corresponding user information, and publishing metadata; a propagation analysis unit for extracting user interaction behaviors involved in cyberbullying incidents based on evidence data to construct a propagation path diagram, and expanding user friend relationships based on the user set in the propagation path diagram to construct a social network subgraph focusing on cyberbullying incidents; a social influence quantification unit for quantifying the social influence of each user in the propagation of cyberbullying incidents based on the propagation path diagram and the social network subgraph; a direct responsibility quantification unit for quantifying the direct responsibility of each user in cyberbullying incidents based on the scale of cyberbullying incidents and the user's offensive language, participation intensity, and participation duration; and a total responsibility quantification unit for comprehensively considering the user's social influence and direct responsibility to obtain the user's total responsibility in cyberbullying incidents.

[0019] For the beneficial effects of the second aspect mentioned above, please refer to the description of the first aspect mentioned above. Attached Figure Description

[0020] Figure 1 A flowchart illustrating a method for quantifying responsibility for cyberbullying incidents, provided as an embodiment of the present invention;

[0021] Figure 2 A schematic diagram of a device for quantifying responsibility for cyberbullying incidents provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of an electronic device structure provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed following the word and its equivalents, but do not exclude other elements or objects.

[0024] This embodiment provides a method and apparatus for quantifying responsibility for cyberbullying incidents.

[0025] See the instruction manual appendix Figure 1Methods for quantifying responsibility for cyberbullying incidents include:

[0026] S101: Obtain evidence data related to cyberbullying incidents, including posting and interaction data related to cyberbullying incidents, corresponding user information, and posting metadata.

[0027] In one possible embodiment, obtaining evidence data for cyberbullying incidents includes: retrieving relevant social data of cyberbullying incidents from evidence data obtained using blockchain evidence storage technology according to set search conditions; the set search conditions are to use strong features of cyberbullying incidents for precise retrieval, and the strong features include the identity identifiers of the participants in the cyberbullying incident, topic identifiers, key account sets, and time ranges; when strong features are insufficient, key entities and time ranges are used as the main search criteria, and the information dissemination chain is expanded by combining social interaction behavior.

[0028] In one specific embodiment, blockchain evidence storage technology is used as the underlying infrastructure and deeply integrated into the data storage architecture of the social platform. This enables user data to be reliably and permanently stored in associated storage at the same time it is generated, providing tamper-proof original data for subsequent accountability.

[0029] Specifically, the evidence-gathering data obtained using blockchain evidence-gathering technology refers to data obtained by collaboratively storing social platform data on-chain and off-chain using blockchain evidence-gathering technology. Off-chain storage includes storing all user social behavior data (including posts, comments, reposts, likes, and private messages) and their metadata (posting time, IP address, device fingerprint, etc.) in the platform's high-efficiency off-chain database (such as a distributed cloud storage system) to meet the business needs of massive data storage and high-speed read / write. On-chain evidence-gathering includes: when a new piece of social data (such as a post or comment) is created and stored in the off-chain database, the hash value of the social data is calculated in real time (such as using the SHA-256 algorithm), and the hash value, creation timestamp, and corresponding user anonymized digital identity are packaged together to form an evidence-gathering transaction and written to the blockchain (which can be a public blockchain, a consortium blockchain, or an evidence-gathering blockchain designed specifically for judicial evidence preservation). The on-chain evidence-gathering process ensures that the "digital fingerprint" of the data is permanently, publicly, and immutably recorded the moment it is created.

[0030] In one possible implementation, users use an anonymous identity for daily activities on social media platforms, while real-name authentication is required during registration. The social media platform's system binds the user's real identity information to the anonymous identity, and this binding relationship is encrypted and stored in a highly secure domain within the platform. This achieves a mechanism where users are anonymous in the foreground but real-name verified in the background. When accountability is required, this binding relationship can be retrieved through a legitimate authorization process to associate the user's anonymous identity with their real identity information.

[0031] In one specific embodiment, after obtaining permission to collect evidence on cyberbullying incidents, relevant social behavior data is retrieved in batches from the off-chain storage data of social media platforms based on the characteristics of the cyberbullying incidents. To address the problem of uncertain retrieval criteria due to the diverse characteristics of cyberbullying incidents, an event profiling + priority rule is adopted to set retrieval conditions. The specific retrieval conditions are: prioritizing the use of strong features of cyberbullying incidents as the primary retrieval criteria for precise retrieval. These strong features include, but are not limited to: the identity identifier (such as ID or URL) of the post / comment / forward object, topic ID, key account set, and a specific time range. Among them, the key account set refers to accounts that are directly related to the occurrence and spread of cyberbullying incidents and constitute key nodes in the event chain, including but not limited to the victim's account, the initial / organizing account, the main perpetrator's voice account, and the main... The search targets accounts involved in the dissemination of information and their associated accounts identified through interactions. When strong features are insufficient, keywords / entities (including variants) plus a time range are used as the primary search criteria. This can be supplemented by social interactions such as forwarding, commenting, and mentioning to extend the information chain. Keywords / entities (including variants) refer not only to the keywords / entities themselves but also to their equivalent or approximate expressions on the platform, including but not limited to aliases / nicknames, abbreviations, homophones, misspellings, pinyin / English spellings, and circumvention techniques such as symbol insertion and space splitting, to retrieve as comprehensive an informational resource as possible related to cyberbullying incidents. The relevant social behavior data retrieved based on the above search criteria constitutes the raw data for evidence collection regarding cyberbullying incidents.

[0032] After obtaining the raw data for evidence collection regarding cyberbullying incidents, the hash value of each piece of raw data is recalculated and compared with the hash value stored on-chain using blockchain evidence storage technology to verify whether the data has been tampered with. If the recalculated hash value matches the on-chain stored hash value, the data has not been tampered with, and the verification passes. The verified data undergoes cleaning and formatting to form standardized, structured evidence data for cyberbullying incidents. Data cleaning includes: merging / deduplicating duplicate data, standardizing time fields, normalizing text encoding and symbols, and deleting or editing status markers. The formatted evidence data is uniformly structured records, specifically including: content (original text / standard text), subject identifier (anonymous identity representation), time information, behavior type (posting / commenting / forwarding, etc.), relationship identifier (such as parent_id / forwarding source), and on-chain evidence storage information (hash / transaction number / verification result) to support subsequent dissemination analysis and responsibility quantification calculations.

[0033] S102: Based on the evidence data, extract the user interaction behaviors involved in the cyberbullying incident to construct a propagation path graph. Based on the user set in the propagation path graph, expand the user's friend relationships to construct a social network subgraph focusing on the cyberbullying incident.

[0034] In one possible embodiment, user interaction behaviors involved in cyberbullying incidents are extracted to construct a propagation path graph. The user set in the propagation path graph is then used as a basis to expand user friend relationships to construct a social network subgraph focusing on cyberbullying incidents. This includes: using users involved in cyberbullying incidents from evidence data as nodes, constructing directed edges from the originator to the interactor based on user interaction behaviors, and assigning weights to the edges according to the nature of the interaction to form a propagation path graph; using the user set in the propagation path graph as the base user set, extracting the users' friends from the social relationship graph of the social platform to form the users' friend set; and constructing a social network subgraph based on the base user set and the users' friend set. The extraction of users' friends from the social relationship graph of the social platform includes: extracting users who have a follow or being followed relationship with the user, or users whose historical interaction frequency with the user is greater than a set threshold.

[0035] For example, the construction of the propagation path graph includes: using users involved in cyberbullying incidents from the evidence data as nodes, including all users who posted, commented, forwarded, or liked related infringing content related to cyberbullying incidents. Directed edges are constructed based on the interaction behavior between users, with the direction of the directed edges pointing from the original creator to the interactor. For example, if user B forwards user A's post, then the edge A→B is constructed. Weights are assigned to the edges based on the nature of the interaction behavior. Different types of interaction behaviors correspond to different weights to reflect their differences in influence; typically, forwarding weight > comment weight > liking weight, and the weight value is a preset constant less than 1.

[0036] The construction of the social network subgraph includes: taking the set of user nodes in the propagation path graph as the base user set, expanding the user's first-degree friends, that is, expanding the user's friend set by having direct followers, being followed, or having historical interactions with users in the base user set with a frequency greater than a preset frequent threshold, and extracting corresponding nodes and edges from the social relationship graph of the social platform based on the user's friend set to reconstruct a social network subgraph focusing on cyberbullying events and related social circles.

[0037] In a social network subgraph, all of a user's friends in their friend set have an edge connected to that user node. The weight of each edge is determined by normalizing the historical interaction frequency between the user and their friends (such as the number of chats, mutual comments, and likes). The steps for calculating the edge weight are as follows:

[0038] In the statistics window Internal collection of users With friends Collection of historical interactive events ,in, Indicates the current time. Indicates the duration of the statistical window. The contribution coefficient is set based on the interaction type. Calculate users With friends Strength of the time decay relationship of the interaction: ,in, Indicates user with his friends The strength of the time decay relationship of the interaction Indicates user With friends Historical interaction events, This represents the natural exponential function. Indicates interactive events The time of occurrence, This indicates an attempt at time decay.

[0039] right Saturation compression is used to suppress extreme values ​​caused by high-frequency interactions or top users in a short period of time, so that the relationship strength exhibits marginal decreasing characteristics in the high range. The saturation compression calculation satisfies the following formula: ,in, This represents the strength of the compressed relationship, and its range is (0~1). This represents the preset scale parameter, used to control the saturation level of relationship strength compression. It can be preset based on the platform's data scale and interaction frequency statistics. It represents the smallest positive number that is not divisible by zero.

[0040] Furthermore, to eliminate the incomparability caused by differences in user activity levels, at the user node... Within the friends collection Performing share-based normalization yields the edges in the social network subgraph. Weights: , ,in, Indicates user with his friends The weight of the interaction edges between users can be interpreted as... The relative distribution ratio of the strength of each friend relationship, express For users Friends Collection The nodes within the graph. If data sparsity causes the denominator to be close to 0, a preset default strategy (such as equal distribution or basic weights) can be used to ensure the robustness of graph construction.

[0041] S103: Quantify the social influence of each user in the spread of cyberbullying incidents based on the propagation path map and social network subgraph.

[0042] In one possible embodiment, for a user, the set of users reachable from the user as the starting point in the propagation path graph is defined as the user's propagation influence set, and the union of the user's propagation influence set and the set of friends forms the user's influence user set.

[0043] The social influence of each user in the spread of cyberbullying events is quantified based on the propagation path diagram and social network subgraph, including: constructing a social feature vector based on the user's social attributes and historical indicators; quantifying the user's social characteristics based on the user's participation in the cyberbullying event and their influence on users in the affected user set; calculating the user's counterfactual behavior prediction difference using the user's social feature vector and social characteristics; and weighting and summarizing the user's influence on each user node in the friend set based on the social network subgraph and the counterfactual behavior prediction difference to obtain the user's friend influence; setting a fixed basic influence weight; and weighting and summarizing the user's influence on each non-friend user node in the affected user set based on the user's participation in the cyberbullying event and the basic influence weight to obtain the user's non-friend influence; and calculating the user's social influence by combining friend influence and non-friend influence.

[0044] The calculation of a user's social characteristics follows the formula: ,in, Indicates target user social characteristics, Indicates user For target users Friends Collection The nodes in Indicates user Participation in cyberbullying incidents Indicates user For users The edge weights in a social network subgraph; the calculation of the user's counterfactual behavior prediction difference satisfies the following formula: ,in, Indicates user The difference in counterfactual behavior prediction, Indicates user Social feature vectors, Indicates user In social characteristics Predicted participation Indicates user Counterfactual predictions of engagement when the influence of social context is removed.

[0045] The user's friend influence is calculated according to the following formula: ,in, Indicates user The influence of friends, Indicates user For users Friends Collection The nodes in Indicates user For users Edge weights in a social network subgraph Indicates user The difference in counterfactual behavior prediction, Indicates user The number of friends in a social network subgraph.

[0046] The user's influence beyond friends is calculated according to the following formula: ,in, Indicates user Non-friend influence, Indicates user For users The impact is on non-friend users in the user set. This indicates the set base influence weight. Indicates user Involvement in cyberbullying incidents.

[0047] In one specific embodiment, the set of user propagation influences is denoted as... The propagation path diagram is based on nodes. User nodes connected by directed edges originating from a given point belong to this set.

[0048] Quantify the indirect liability of infringing users in the spread of events due to their social influence, and define this indirect liability as the user's social influence SI. The user's SI is determined by their contribution to the behavior of the user set U, as follows: The user set is determined by the user's influence. The propagation impact set and the friend set are jointly determined, and duplicates are removed using the union method: This avoids the same user being counted twice because they belong to both the propagation chain and their friend relationships. A user's social influence is calculated according to the following formula: ,in, Indicating influence, this refers to the aggregate of users across the set U of users who influence others. The impact on other users through social relationships, specifically in U and users In a social network subgraph, user nodes (i.e., friend nodes) with social edges are weighted using edge weights. The intensity of the influence of the characterization; This refers to the influence of non-friends, specifically the users aggregated across the set U of users influencing the user group. The fundamental impact on non-friend nodes, specifically on U and users. User nodes that do not have a direct social edge (i.e., stranger nodes) have an influence based on a fixed base influence weight. The overall participation of users at this node in cyberbullying incidents The product representation; and The preset weighting coefficients, and ,generally This is reflected in the information dissemination process, where individuals have a much higher level of trust, identification, and attention towards statements from friends than from strangers. Therefore, the influence of friends is more effective, and the users they influence are more numerous. The indirect responsibilities they should bear are also correspondingly greater.

[0049] In one possible implementation, for cyberbullying incidents, statistics are displayed within a window based on user... The data on interactive behaviors (including posting, forwarding, commenting, liking, etc.) is used to calculate the overall engagement level. The specific steps for calculating overall participation are as follows:

[0050] In the evidence data of cyberbullying incidents, statistics on users The number of valid occurrences (or frequencies) of various behaviors within the statistical window. ,in , Indicates the type of behavior. This refers to the act of posting. Indicates forwarding behavior, Indicates the act of commenting. This indicates the act of giving a "like".

[0051] To reflect the different contributions of various behaviors to propagation and diffusion, weighting coefficients are assigned to different behavior types. Preset constants or configuration items to obtain unnormalized participation strength: ,in, Users representing the aggregated data The level of participation in various behaviors within the statistics window. Indicates behavior The weighting coefficients, Indicates behavior The effective number of occurrences.

[0052] To ensure cross-user comparability, The final overall engagement score is obtained by normalizing the score within the set of affected users. ,in, Indicates user Overall participation express To influence user sets Nodes within, It represents the smallest positive number that is not divisible by zero.

[0053] In one specific embodiment, the calculation of friend influence includes:

[0054] For each user Constructing feature vectors, specifically including the user's own social feature vectors. Social characteristics Among them, social feature vectors It is composed of user profiles and historical behavior logs extracted from social media platforms, along with user attributes and historical statistical indicators, including the number of followers, the number of people following, the number of mutual follows, historical activity level, and account registration duration. To eliminate differences in the units of measurement of different indicators and suppress the influence of extreme values, each indicator is standardized and then concatenated to form the final result. . By aggregating variables of users in the friend set Calculations show that The calculation satisfies the following formula: ,in, Indicates target user social characteristics, Indicates user For target users Friends Collection The nodes in Indicates user Participation in cyberbullying incidents Indicates user For users Edge weights in a social network subgraph.

[0055] Train a prediction function using a machine learning model (such as a neural network, gradient boosting tree, etc.). The prediction function is trained using a pre-defined training sample set: for each sample user... Take its feature vector [[] as input, based on its actual observations of the overall participation rate] As a supervisory signal (label), the model parameters are iteratively optimized to minimize the error between the predicted and true values, thereby learning the mapping relationship between the user's social environment and their participation behavior. The prediction function is defined as follows: , This indicates that the prediction function is applied to a given user's social feature vector. Social characteristics Under these conditions, the predicted value of overall user engagement.

[0056] For users in the affected user set Calculate the difference in its counterfactual behavior predictions. The calculation formula is as follows: ,in Indicates user In social characteristics Predicted engagement, i.e., user engagement Predicted value ; This represents the counterfactual prediction engagement when social characteristics are set to zero to remove the influence of the social environment. Counterfactual behavioral prediction difference. Used to measure users Increased behavioral engagement due to its specific social context, i.e., the degree to which it is affected.

[0057] Based on the social network subgraph and the counterfactual behavior prediction difference calculated above, the influence is assessed on each user node in the user set U. The contributions are weighted and aggregated to obtain the user's contribution. Friends influence The calculation formula is as follows: ,in, Indicates user The influence of friends, Indicates user For users Friends Collection The nodes in Indicates user For users Edge weights in a social network subgraph Indicates user The difference in counterfactual behavior prediction, Indicates user The number of friends in a social network subgraph. Indicates user The proportion influenced by one of the friends, if A value of 1 indicates that the influence of friends is 100%. Quantitatively reflects the user The indirect impact of the spread of an event through networks of friends.

[0058] In a specific embodiment, the user's non-friend influence is calculated according to the following formula: ,in, Indicates user Non-friend influence, Indicates user For users The impact is on non-friend users in the user set. This indicates the set base influence weight. Indicates user Involvement in cyberbullying incidents.

[0059] According to the formula The user's social influence is obtained by combining the influence of friends and non-friends.

[0060] S104: Quantify the direct responsibility of each user in cyberbullying incidents based on the scale of the incident and the user's offensiveness, intensity of participation, and duration of participation.

[0061] In one possible embodiment, the direct responsibility of each user in a cyberbullying incident is quantified based on the scale of the incident and the user's offensive language, participation intensity, and duration of participation. This includes: classifying the scale of a cyberbullying incident based on the number of participating users and the total amount of related content to obtain an incident scale coefficient; using a pre-trained semantic analysis model to identify the offensiveness of user-posted text content in the evidence data to obtain an offensiveness score; calculating the user's offensive language score based on the offensiveness scores of all user-posted text content; obtaining the user's participation intensity based on the number of user-posted text content with offensiveness scores greater than a set threshold; calculating the time span of user participation in the cyberbullying incident; and normalizing the incident scale coefficient, offensive language score, participation intensity, and time span, then performing a weighted sum to obtain a quantified value of the user's direct responsibility in the cyberbullying incident.

[0062] In one specific implementation, direct responsibility is used to measure the direct harm of a user's own words and actions. When quantifying direct responsibility, the user's relevant behaviors in cyberbullying incidents are quantitatively assessed from multiple dimensions, and the indicators of each dimension are normalized to eliminate the influence of unit of measurement. The dimensions for quantifying direct responsibility include the event scale coefficient, the degree of verbal aggression, the intensity of participation, and the duration of participation.

[0063] Among them, the event scale coefficient reflects the background impact of the overall event scale on the individual responsibility assessment: based on the total number of users involved in the cyberbullying event. Total amount of content related to the event The system categorizes incidents into different levels and assigns corresponding coefficients to each level. For example, a tiered threshold rule is used to obtain the event scale coefficient for cyberbullying incidents. Small: and ,Pick Medium: Meets the requirements or ,Pick Large: meets the requirements or ,Pick . and These are preset parameters, which can be calibrated based on platform governance rules or historical event statistics.

[0064] Using pre-trained NLP sentiment analysis or violent speech recognition models, we perform aggressiveness identification on each piece of text content posted by users (including posts, comments, and private messages), targeting user... For each text k, output the attack score of that text. This is used to characterize the degree to which it contains aggressive elements such as insult, defamation, and threats. Let... For users The user's verbal aggression score is determined by the set of event-related texts. for: , To prevent extremely small positive numbers from being divided by zero.

[0065] The intensity of user participation is determined by the number of infringing content posted by users in cyberbullying incidents. Infringing content includes posts, comments, and private messages deemed to contain offensive elements such as insults, defamation, and threats; that is, the offensiveness score of the text exceeds a set offensive threshold. The number of contents is calculated to avoid including irrelevant content.

[0066] Calculate the time span from a user's first involvement in cyberbullying to their last involvement in cyberbullying. The unit for this time span is optional and can be one of seconds, minutes, or hours. ,in, Indicates user The timestamp of the last action in content related to cyberbullying incidents. Indicates user The first action timestamp in content related to cyberbullying incidents.

[0067] After normalizing the event scale coefficient, verbal aggression score, participation intensity, and time span, the quantified value of the user's direct responsibility (DR) is calculated using a weighted summation formula. , , , , These are the weighting coefficients for each dimension, and their values ​​can be adjusted according to the actual situation.

[0068] S105: The user's overall responsibility in cyberbullying incidents is determined by combining the user's social influence and direct responsibility.

[0069] user The total liability is the sum of its direct and indirect liabilities: .

[0070] In one possible embodiment, after obtaining the user's total responsibility in the cyberbullying incident, the method further includes: sorting the total responsibility quantification values ​​of all users who participated in the cyberbullying incident to generate a macro responsibility quantification ranking report; and generating an individual tort liability detail report based on the detailed information related to the direct and indirect responsibilities of each user who participated in the cyberbullying incident.

[0071] In one specific implementation, all infringing users are traversed, their TR (Traffic Response) values ​​are calculated, and they are sorted to generate two reports: A "Macro-Level Liability Quantification Ranking Report": This report displays the TR value rankings of all infringing users in list or chart format, helping regulatory agencies and platforms quickly identify major infringers and determine priority for action. An "Individual Infringement Liability Details Report": This report details the DR (Depth of Responsibility) value (and scores for each dimension), SI (Score of Infringement), and their calculation details for each individual infringing user, forming a complete chain of evidence. This report can be directly used as supporting evidence for complaints to the platform or lawsuits in court.

[0072] The method for quantifying responsibility for cyberbullying incidents provided by this invention deeply integrates blockchain evidence storage technology into the underlying layer of social platforms, enabling simultaneous data generation and credible solidification. This method ensures the originality, integrity, and immutability of all electronic evidence, fundamentally solving the pain points of electronic evidence being easily lost and altered, and greatly enhancing the credibility and probative value of evidence in judicial proceedings.

[0073] An innovative dual-track quantitative model of "direct responsibility + indirect responsibility" is adopted. In calculating direct responsibility, multi-dimensional indicators such as NLP and behavioral statistics are comprehensively used to assess the direct harm of the infringement. In calculating indirect responsibility, a causal inference model based on synthetic control and social influence calculation is introduced. By constructing counterfactual scenarios and using machine learning predictions, correlation and causality can be separated, accurately quantifying the true contribution of each user to the spread of the event due to their network influence. This avoids the drawbacks of rough assessments based solely on network topology (such as centrality indicators) or single forwarding volume, making the final liability determination more convincing and fair.

[0074] The system automates the entire process from data collection, consolidation, analysis to report generation. Victims no longer need to manually perform massive and tedious evidence gathering and notarization work; the system's one-click generation of macro and individual liability reports provides them with powerful tools for protecting their rights. Simultaneously, the reports offer platforms and regulatory agencies clear priorities and detailed criteria for penalties, significantly improving governance efficiency and enabling rapid and accurate responses to cyberbullying incidents.

[0075] Compared to the difficult-to-explain "black box" models or traditional methods that only reflect correlations, the SI calculation of this invention is based on counterfactual reasoning, and its calculation results (social value) have clear behavioral implications (e.g., user A's influence directly led to his fan group increasing infringing behavior by X times). This interpretability makes the liability determination process more transparent, and the conclusions are more easily understood and accepted by the parties involved, platforms, and judicial institutions.

[0076] The responsibility of any participant in cyberbullying incidents, whether a direct initiator or an indirect facilitator, can be clearly quantified and traced. This powerful ability to trace and trace such incidents effectively deters potential perpetrators, helps curb the growth and spread of cyberbullying, and fosters a cleaner, more civilized, and healthier online environment.

[0077] See the instruction manual appendix Figure 2 This embodiment also provides a device for quantifying responsibility for cyberbullying incidents, which is used to implement the above-described method embodiment. The device includes:

[0078] The evidence collection unit 201 is used to obtain evidence data of cyberbullying incidents, including posting and interaction data of cyberbullying incidents, corresponding user information and publishing metadata.

[0079] The propagation analysis unit 202 is used to extract user interaction behaviors involved in cyberbullying incidents based on evidence data to construct a propagation path graph. Based on the user set in the propagation path graph, the user's friend relationships are expanded to construct a social network subgraph focusing on cyberbullying incidents.

[0080] The social influence quantification unit 203 is used to quantify the social influence of each user in the spread of cyberbullying incidents based on the propagation path graph and social network subgraph.

[0081] The direct responsibility quantification unit 204 is used to quantify the direct responsibility of each user in a cyberbullying incident based on the scale of the incident and the user's offensive language, intensity of participation, and duration of participation.

[0082] The total responsibility quantification unit 205 is used to calculate the user's total responsibility in cyberbullying incidents by combining the user's social influence and direct responsibility.

[0083] All relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0084] In other embodiments of this application, an electronic device is disclosed, such as... Figure 3As shown, the electronic device 300 may include: one or more processors 301; a memory 302; a display 303; one or more application programs (not shown); and one or more computer programs 304. These devices can be connected via one or more communication buses 305. The one or more computer programs 304 are stored in the memory and configured to be executed by the one or more processors 301. The one or more computer programs 304 include instructions that can be used to perform actions such as... Figure 1 And the various steps in the corresponding embodiments.

[0085] Through the above description of the embodiments, those skilled in the art will clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0086] In the embodiments of this application, the functional units can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0087] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, essentially, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as flash memory, portable hard disk, read-only memory, random access memory, magnetic disk, or optical disk.

[0088] The above description is merely a specific implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the embodiments of this application should be covered within the protection scope of the embodiments of this application. Therefore, the protection scope of the embodiments of this application should be determined by the protection scope of the claims.

Claims

1. A method for quantifying the responsibility of a network violence event, characterized by, The method comprises the following steps: acquiring forensic data of a network violence event, including post interaction data of the network violence event, corresponding user information and publishing metadata; constructing a propagation path graph based on user interaction behavior of the network violence event according to the forensic data, expanding a friend relationship of a user based on a user set in the propagation path graph to construct a social network subgraph focusing on the network violence event; quantifying social influence of each user in the propagation of the network violence event according to the propagation path graph and the social network subgraph; quantifying direct responsibility of each user in the network violence event according to a size of the network violence event and speech aggressiveness, participation intensity and participation duration of the user; comprehensively quantifying total responsibility of the user in the network violence event based on the social influence and the direct responsibility of the user.

2. The method of claim 1, wherein, The acquiring of the forensic data of the network violence event comprises the following steps: retrieving relevant social data of the network violence event from stored data obtained by using a blockchain storage technology according to set retrieval conditions; the set retrieval conditions are used to accurately retrieve the network violence event by using strong features of the network violence event, the strong features including an identity of a participant of the network violence event, a topic identifier, a key account set and a time range; when the strong features are insufficient, key entities and the time range are used as main retrieval bases, and information propagation link expansion is performed in combination with social interaction behavior.

3. The method of claim 1, wherein, The extracting of the user interaction behavior of the network violence event and the constructing of the propagation path graph based on the user set in the propagation path graph to construct the social network subgraph focusing on the network violence event comprise the following steps: taking the user participating in the network violence event in the forensic data as a node, constructing a directed edge from an originator to an interactor according to interaction behavior between the users, and assigning a weight to the edge according to an interaction nature to form the propagation path graph; taking the user set in the propagation path graph as a basic user set, extracting friends of the user from a social relationship graph of a social platform to form a friend set of the user, and constructing the social network subgraph according to the basic user set and the friend set of the user; wherein the extracting of the friends of the user from the social relationship graph of the social platform comprises: extracting a user having a follow or followed relationship with the user, or a user having a historical interaction frequency greater than a set threshold with the user.

4. The method of claim 3, wherein, for a user, defining a user set reachable from the user as a starting point in the propagation path graph as a propagation influence set of the user, and forming an influence user set of the user by taking a union set of the propagation influence set and the friend set of the user; the quantifying of the social influence of each user in the propagation of the network violence event according to the propagation path graph and the social network subgraph comprises the following steps: constructing a social feature vector of the user according to a social attribute and a historical index of the user, quantifying the social feature of the user according to a participation degree of the user in the network violence event and an influence on the user in the influence user set, calculating a counterfactual behavior prediction difference of the user by using the social feature vector and the social feature of the user, and obtaining a friend influence of the user by using the social network subgraph and the counterfactual behavior prediction difference to weight and sum up influences of the user on each user node in the friend set of the user. The fixed basic influence weight is set, the influence of each user node in the influence user set by the user who is not a friend is weighted and summarized based on the participation of the user in the network violence event and the basic influence weight, and the non-friend influence of the user is obtained; The social influence of the user is calculated by integrating the friend influence and the non-friend influence.

5. The method of claim 4, wherein, The calculation of a user's social characteristics follows the formula: ,in, Indicates target user social characteristics, Indicates user For target users Friends Collection The nodes in Indicates user Participation in cyberbullying incidents Indicates user For users Edge weights in a social network subgraph; The user's counterfactual behavior prediction difference value satisfies the following formula: wherein, represents a counterfactual behavior prediction difference value of the user , represents a social feature vector of the user , represents a predicted participation degree of the user under the social feature , represents a counterfactual predicted participation degree of the user when the social environment influence is removed.

6. The method of claim 4, wherein, The friend influence of a user satisfies the following formula: wherein denotes the friend influence of a user , denotes the number of nodes in the friend set of a user , denotes a node in the friend set of a user , denotes the edge weight of a user to a user in the social network subgraph, denotes the counterfactual behavior prediction difference of a user , denotes the number of friends of a user in the social network subgraph.

7. The method of claim 4, wherein, The non-friend influence of a user satisfies the following formula: wherein represents the non-friend influence of a user represents the influence of a user on a non-friend in the influence user set of the user represents a set influence weight, represents the participation of a user in a network violence event.​​ 8. The method of claim 1, wherein, The direct responsibility of each user in the network violence event is quantified according to the size of the network violence event and the speech aggressiveness, participation intensity and participation duration of the user, including: The size coefficient of the network violence event is obtained according to the number of participating users and the total amount of related content of the network violence event; The aggressiveness score of the user is calculated according to the aggressiveness score of the text content published by all users by using the pre-trained semantic analysis model to identify the aggressiveness of the text content published by the user in the evidence data; The participation intensity of the user is obtained according to the number of content in the text content published by the user whose aggressiveness score is greater than a set threshold; The time span of the user participating in the network violence event is calculated; The event size coefficient, the speech aggressiveness score, the participation intensity and the time span are normalized and then weighted and summed to obtain the quantified value of the direct responsibility of the user in the network violence event.

9. The method of claim 1, wherein, The total responsibility of the user in the network violence event is the sum of the direct responsibility and the indirect responsibility; After obtaining the total responsibility of the user in the network violence event, it further includes: The total responsibility quantified values of all users participating in the network violence event are sorted to generate a macro responsibility quantified ranking report; An individual tort liability detail report is generated according to the direct responsibility and the indirect responsibility related detail information of each user participating in the network violence event.

10. A network violence event liability quantification apparatus characterized by, The device includes: An evidence collection unit configured to collect evidence data of a network violence event, including post interaction data, corresponding user information and publishing metadata of the network violence event; A propagation analysis unit configured to extract user interaction behaviors participating in the network violence event according to the evidence data to construct a propagation path graph, and to expand the friend relationship of the user based on the user set in the propagation path graph to construct a social network subgraph focusing on the network violence event; A social influence quantification unit configured to quantify the social influence of each user in the propagation of the network violence event according to the propagation path graph and the social network subgraph; A direct responsibility quantification unit configured to quantify the direct responsibility of each user in the network violence event according to the size of the network violence event and the speech aggressiveness, participation intensity and participation duration of the user; A total responsibility quantification unit configured to integrate the social influence and the direct responsibility of the user to obtain the total responsibility of the user in the network violence event.

Images