A system for preventing obscenity to minors

Abstract

The application discloses a kind of prevent juvenile obscenity governance system, belong to network security and juvenile protection field.The system includes multi-source data acquisition module, identity verification and juvenile identification module, criminal script behavior analysis engine, risk assessment and early warning module, intervention execution module and electronic evidence solidification module.System based on criminal script theory will air obscene crime deconstruction into preparation, contact, violation and maintain four stages, through behavior sequence model to the multi-platform user interactive data time series analysis, in combination with multidimensional risk assessment index system calculates risk score and executes hierarchical intervention, simultaneously using block chain storage technology automatically solidifies electronic evidence.The application realizes from front-end prevention to rear-end repair whole chain governance, effectively solve the problem that traditional system cannot identify progressive induction behavior and electronic evidence solidification difficulty.

Description

Technical Field

[0001] This invention relates to cybersecurity and the protection of minors, and more particularly to a governance system for preventing molestation of minors. Background Technology

[0002] With the rapid development of internet technology and the widespread use of smart devices, minors have become frequent users of cyberspace. However, the anonymity and cross-regional nature of the internet have also facilitated illegal and criminal activities targeting minors. In recent years, "remote molestation" has become a prominent form of crime infringing upon the rights of minors. This involves perpetrators using online media to coerce or induce minors to send private photos or engage in nude chats through verbal threats and emotional manipulation. Although such behavior involves no physical contact, the psychological trauma and privacy risks inflicted on minors are more lasting and severe.

[0004] Currently, the governance of remote indecent assault crimes faces the following main challenges: First, the crimes are characterized by their phased and covert nature, making traditional post-incident punishment models ineffective in prevention; second, the regulatory responsibilities of online platforms are not being fully implemented, and these platforms, as key hubs for the occurrence of crimes, lack effective technical monitoring methods; third, obtaining electronic evidence is difficult, as perpetrators use virtual identities and encrypted communication, increasing the difficulty of tracing the source, and once private materials are disseminated online, they are difficult to completely erase; fourth, there is a lack of coordination mechanisms among various governance entities, with serious information silos between the judiciary, families, schools, and platforms.

[0005] In existing technologies, some content moderation systems rely solely on keyword filtering or single-dimensional image recognition, failing to effectively identify remote indecent acts with complex rhetoric and gradual inducement. Some child protection programs focus only on usage time limits or content ratings, lacking systematic modeling and targeted intervention capabilities across the entire chain of criminal behavior. Therefore, a comprehensive governance system capable of identifying, warning, and intervening from the perspective of the entire criminal process is urgently needed. Summary of the Invention

[0006] Purpose of the invention: The purpose of this invention is to provide a governance system and method for preventing child molestation. Based on the crime script theory, the invention deconstructs remote molestation crimes into four stages: preparation, contact, infringement, and maintenance. For each stage, a behavior recognition model and intervention strategy are established to achieve full-chain governance from front-end prevention to back-end repair.

[0007] Technical Solution: A governance system for preventing child molestation, comprising: a multi-source data acquisition module, which collects user interaction data in real time from at least two internet platforms, including text messages, image information, voice information, video information, user registration information, login behavior data, and social relationship chain data; an identity verification and minor identification module, which performs dual verification of real-name authentication and facial liveness detection on registered users, constructs a user age profile model based on user registration information, behavioral feature data, and device fingerprint information to determine whether the user is a minor, and marks the accounts of identified minor users with protection tags; and a crime script behavior analysis engine, which constructs a behavior sequence model including preparation, contact, infringement, and maintenance phases based on crime script theory, extracts temporal behavioral features from the user interaction data, and matches the extracted behavioral features with the behavior sequence model. The system includes a matching module to identify the current stage of the crime script; a risk assessment and early warning module that calculates a risk score based on the stage matching results output by the crime script behavior analysis engine and a preset risk assessment index system. When the risk score exceeds a preset threshold, it generates a corresponding level of early warning signal based on the current stage of the crime script and pushes it to the intervention execution module; the intervention execution module executes graded intervention measures based on the level of the early warning signal, including pushing risk warnings to underage users, restricting the behavior of suspected accounts, sending alarm notifications to guardians, and pushing case clues to law enforcement agencies; and an electronic evidence solidification module that automatically calculates the hash value of relevant interaction data, user identity information, and behavior time chain when it detects interaction behavior that matches the characteristics of a crime script, writes the obtained hash value into the blockchain for storage, and forms an electronic evidence chain with timestamps and tamper-proof attributes.

[0008] Furthermore, the user age profile model in the identity verification and minor identification module adopts a multilayer perceptron network structure. The input layer receives user registration time, active period distribution, language style feature vector, and social relationship graph features. The output layer outputs the probability value that the user is a minor. When the probability value is greater than the preset age determination threshold, the user is marked as a suspected minor and a further identity verification process is triggered.

[0009] Furthermore, the behavioral sequence model sets identification rules for each stage: the preparation stage detects whether the account is newly registered, frequently switches devices, or uses a virtual identity; the contact stage detects whether adult users initiate excessive message interactions with underage users or use emotionally manipulative language; the infringement stage detects whether the interaction content contains features such as soliciting private pictures or videos, threats, or intimidation; and the maintenance stage detects whether there is a behavioral pattern of continuously blackmailing or spreading private materials across platforms.

[0010] Furthermore, the risk assessment indicator system includes five indicators: abnormality of interaction frequency, sensitivity of content, historical risk record of the actor, vulnerability of the victim's age, and correlation of cross-platform behavior. The risk score R = Σ(wi×Si), where Si is the normalized score value of the i-th item, wi is the weight, and the sum of the weights is 1. When R is less than the first threshold, it is an observation-level warning; when it is between the first and second thresholds, it is an alert-level warning; and when it is greater than or equal to the second threshold, it is an emergency-level warning.

[0011] Furthermore, when writing to the blockchain, the electronic evidence solidification module uses the original content of the interaction data, the user identifiers of the sender and receiver, the message timestamp, and the device information as the data to be stored. It uses the SHA-256 algorithm to calculate the hash digest value of the data to be stored, packages the hash digest value together with the storage timestamp into a storage transaction record and writes it to the blockchain network. It also organizes multiple storage records through a Merkle tree structure to form a traceable and verifiable evidence chain.

[0012] Furthermore, the intervention execution module implements tiered intervention measures according to the warning level: when an observation-level warning is triggered, a safety risk education prompt is pushed to the underage user's terminal and the granularity of log recording is increased; when a warning-level warning is triggered, the message sending frequency and friend-adding function of the suspected account are restricted, and an alarm notification is sent to the guardian; when an emergency-level warning is triggered, the suspected account is immediately frozen and all session data is saved, and case clues and solidified electronic evidence are simultaneously pushed to the law enforcement agency's information system.

[0013] Furthermore, it also includes a cross-platform collaborative governance module, which connects with the risk control systems of multiple Internet platforms through standardized interfaces to establish a cross-platform user risk profile sharing mechanism. When a platform identifies a high-risk user, it synchronizes the risk identifier and feature summary of the high-risk user to the cross-platform collaborative governance module, which then pushes the risk identifier to other connected platforms to achieve cross-platform collaborative risk prevention and control.

[0014] Furthermore, it also includes a victim protection and assistance module, which automatically initiates the victim protection process after the case is confirmed. The protection process includes: triggering a full-network investigation and deletion command for privacy information, searching and deleting relevant privacy content on various docking platforms through hash value matching; generating a victim psychological assistance referral form and pushing it to the docking psychological counseling institution; establishing a victim protection file to record the implementation of assistance measures and regular follow-up information.

[0015] This invention also provides a method for preventing child molestation, applied to the aforementioned governance system, comprising the following steps: S1. Collect user interaction data in real time from at least two Internet platforms, including text messages, image information, voice information, video information, user registration information, login behavior data, and social relationship chain data; S2. Perform dual verification of real-name authentication and face liveness detection on registered users. Based on user registration information, behavioral characteristics and device fingerprints, build a user age profile model to determine whether the user is a minor and mark the account of underage users with a protection label. S3. Based on the crime script theory, construct a behavioral sequence model that includes four stages: preparation, contact, infringement and maintenance. Extract temporal behavioral features from user interaction data, match the extracted behavioral features with the identification rules of each stage, and identify the crime script stage in which the current interaction behavior is located. S4. Based on the stage matching results, calculate the risk score R=Σ(wi×Si) using an indicator system that includes anomaly of interaction frequency, sensitivity of content, historical risk record of the actor, vulnerability of the victim's age, and cross-platform behavioral correlation. S5. When the risk score exceeds the preset threshold, a corresponding level warning signal is generated based on the current stage of the crime script, and graded intervention measures are implemented according to the level, including risk warning push, restriction of suspected accounts, warning to guardians, and push of case clues to law enforcement agencies. S6. When an interaction behavior that matches the characteristics of a criminal script is detected, the hash value of the interaction data, user identity information and behavior time chain is automatically calculated, and the hash value is written into the blockchain for storage, forming an immutable electronic evidence chain.

[0016] Beneficial effects: (1) Based on the crime script theory, a behavior sequence model is constructed, which can systematically identify the behavioral characteristics of remote molestation crimes from preparation to maintenance. This breaks through the limitation of traditional systems that can only perform single-point detection and achieves accurate identification of the entire crime chain. (2) A multi-dimensional risk assessment indicator system is adopted for quantitative scoring. Combined with a graded early warning and graded intervention mechanism, appropriate intervention measures can be automatically matched according to the degree of risk, thus realizing a precise and differentiated governance strategy. (3) By using blockchain evidence storage technology to automatically solidify electronic evidence, the integrity and immutability of the evidence chain are ensured, effectively solving the problems of difficulty in obtaining and solidifying electronic evidence in remote molestation cases; (4) Through the cross-platform collaborative governance mechanism and the victim protection and assistance module, a comprehensive governance system with multiple stakeholders including the judiciary, family, school and platform has been realized. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the overall architecture of the system for preventing molestation of minors according to the present invention.

[0018] Figure 2 This is a flowchart of the method for preventing molestation of minors according to the present invention.

[0019] Figure 3 This is a diagram showing the correspondence between the four stages of the crime script and the graded intervention in this invention. Detailed Implementation

[0020] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example 1

[0022] A governance system to prevent child molestation, such as Figure 1 As shown, the governance system comprises the following six core modules: (1) Multi-source data acquisition module: This module connects with multiple internet platforms through standardized API interfaces to collect user interaction data on each platform in real time. The data acquisition scope covers instant messaging platforms, social networking platforms, online game platforms, and live streaming / short video platforms. The collected interaction data includes: text message content and its sending time, metadata information of images and videos, real-name authentication information submitted by users during registration, user login IP address and device identifier, user's friend list and social relationship graph. Data from each platform is uniformly aggregated to the data processing center through message queues.

[0023] (2) Identity Verification and Minor Identification Module: During user registration, users are required to complete dual verification of ID card information submission and facial liveness detection. For users who have passed real-name authentication, the system continuously collects their behavioral characteristic data, including the distribution of active time periods, language style used, social circle characteristics, and device usage habits, to construct a user age profile model. This model adopts a multilayer perceptron network structure. The input layer receives the above behavioral characteristic vectors, the hidden layer has three layers with 128, 64, and 32 nodes respectively, and the output layer outputs a probability value between 0 and 1, representing the probability that the user is a minor. When the probability value exceeds 0.7, the user is marked as a suspected minor, triggering a manual review process or further identity verification. For accounts confirmed to be minors, the system automatically marks them with a protection tag and includes them in the key protection scope.

[0024] (3) Crime script behavior analysis engine: This engine is the core analysis component of the system. Based on the crime script theory, it decomposes remote molestation crimes into four progressive stages and establishes an identification rule base for each stage.

[0025] The identification rules for the preparation stage include: newly registered accounts and frequent switching of device identifiers, use of virtual identities that do not match the registration information, and adding underage users as friends in batches on multiple platforms, among other preparatory behavioral characteristics.

[0026] The identification rules for the contact stage include: adult users initiating message interactions with underage users that exceed the normal frequency of social interaction (such as the number of messages per day exceeding 3 times the average number of interactions for users of that age group), using emotionally manipulative language (such as typical manipulative statements like "Brother will protect you" or "It's a secret that belongs only to us"), and contacting under false identities (such as peers, talent scouts, or psychological counselors).

[0027] The rules for identifying the stage of infringement include: hints or explicit requests for private photos in the interaction content, demands for the other party to expose themselves in videos, and sending content containing threats or intimidation (such as "If you don't listen to me, I will send out your photos").

[0028] The rules for identifying the maintenance phase include behavioral patterns such as using obtained private materials to continuously blackmail victims, spreading and disseminating private materials across different platforms, and exercising long-term psychological control over victims.

[0029] The engine extracts temporal behavioral features from user interaction data and uses a sliding time window mechanism (window size is 7 days) to match the behavioral sequence within the window with the four-stage recognition rule base one by one, and outputs the stage and confidence level of the current behavior.

[0030] (4) Risk Assessment and Early Warning Module: Based on the stage matching results output by the crime script behavior analysis engine, a comprehensive risk score is calculated. The risk assessment indicator system includes five indicators: Interaction frequency anomaly index S1: Calculate the ratio of the message interaction frequency between the suspected user and the underage user to the average interaction frequency of users of the same age group on the platform, and normalize it to the 0-1 range. Content sensitivity index S2: The sensitivity of interactive text and images is scored using a pre-trained content classification model, and a sensitivity value between 0 and 1 is output. Historical Risk Record Indicator S3: Query the suspect user's historical reporting and violation records across the entire platform. No records result in a value of 0, and each additional record increases the value by 0.2, with a maximum of 1. Victim Age Vulnerability Index S4: Values ​​are assigned based on the victim's age: 1.0 for under 12 years old, 0.7 for 12-14 years old, 0.4 for 14-16 years old, and 0.2 for 16-18 years old. Cross-platform behavioral correlation index S5: Detects whether a suspect user initiates interactions with the same underage user on multiple platforms. The index is 0.3 when the number of cross-platform interactions is 1, 0.6 when it is 2, and 1.0 when it is greater than or equal to 3.

[0031] The comprehensive risk score is calculated as follows: R = w1×S1 + w2×S2 + w3×S3 + w4×S4 + w5×S5, where the weighting coefficients are w1=0.20, w2=0.30, w3=0.15, w4=0.20, and w5=0.15, respectively.

[0032] Warning levels are classified as follows: when R < 0.4, it is an observation level warning; when 0.4 ≤ R < 0.7, it is a warning level warning; and when R ≥ 0.7, it is an emergency level warning.

[0033] (5) Intervention Execution Module: Automatically executes tiered interventions based on the warning level. When an observation-level warning is triggered, the system pushes a safety risk education prompt to the underage user's terminal (such as "Please pay attention to protecting your personal privacy and do not send private photos to strangers"), and simultaneously increases the granularity of the session's log recording to save each message individually. When a warning-level warning is triggered, the system restricts the message sending frequency of the suspected account (no more than 5 messages per hour), prohibits the addition of new friends, and sends an alarm notification to the underage user's bound guardian. When an emergency-level warning is triggered, the system immediately freezes the suspected account, saves a snapshot of all session data, and simultaneously pushes the case clues along with the solidified electronic evidence to the connected law enforcement agency's information system.

[0034] (6) Electronic Evidence Solidification Module: When the crime script behavior analysis engine detects interactive behavior that matches the characteristics of any stage of the crime script, it automatically triggers the evidence solidification process. This module extracts the original content of the interactive data, the user identifiers of the sender and receiver, the message timestamp, the sending device and IP address, etc., as data to be stored. It uses the SHA-256 algorithm to calculate the hash digest value of the data to be stored, and packages the hash digest value together with the storage timestamp into a storage transaction record, which is then written into the consortium blockchain network. Multiple storage records are organized through a Merkle tree structure to form a complete evidence chain that can be traced and verified. Law enforcement agencies can verify the integrity of any node in the evidence chain through the verification interface.

[0035] Example 2

[0036] A governance approach to prevent child molestation, such as Figure 2 As shown, the method includes the following steps: Step S1: Real-time Data Acquisition from Multiple Sources. The system establishes real-time data channels with five internet platforms through a data gateway. These platforms include instant messaging platform A, social networking platform B, online game platform C, short video platform D, and learning tool platform E. (Before data collection, a pre-compliance check is performed to verify the compliance of the collected data, using a method of explicit authorization from the guardian of the minor and verification based on the minimum necessary collection rule). The data gateway adopts an Apache Kafka message queue architecture. Each platform pushes user interaction events to the message queue in JSON format, and the system consumes the message data with a second-level delay. Each interaction message contains at least the following fields: unique message identifier, sender user ID, receiver user ID, message type (text / image / voice / video), message content hash value, sending timestamp, and device fingerprint information.

[0037] Step S2: User Identity Verification and Minor Identification. When a new user registers, the system requires submission of photos of both sides of their ID card and performs facial liveness detection. The system uses OCR to identify the birth date in the ID card number to calculate the user's actual age. For users who have passed real-name authentication, the system reassesses the user's age profile every 30 days. The specific parameters of the age profile model are: 64-dimensional input features, including active time distribution features (24 dimensions, representing the proportion of active users within 24 hours), language style features (20 dimensions, extracted through a pre-trained text style embedding model), social relationship graph features (15 dimensions, including the average age distribution of friends, the proportion of minors among friends, and the frequency of newly added friends), and device features (5 dimensions, including device type and operating system version).

[0038] The model was trained on a labeled historical user dataset (containing 100,000 user samples with confirmed ages), using the cross-entropy loss function, with a learning rate of 0.001 and 100 training epochs. The model achieved an AUC of 0.93 and a recall (with a threshold of 0.7) of 0.91 on the test set.

[0039] Step S3: Crime script behavior sequence matching. The system performs behavior sequence analysis on each pair of interaction sessions involving underage users.

[0040] The sliding time window is set to 7 days, and the extracted behavioral sequences within the window include: Session initiation mode: which party initiates the session, the time interval between the first sessions, and the trend of session frequency changes; Evolution of conversational content: The speed and manner of transitioning from normal social topics to sensitive topics; Information solicitation behavior: Whether there are any requests to send photos, videos, or personal information; Characteristics of emotional manipulation: Whether emotional blackmail, threats, intimidation, or material inducements are used.

[0041] The matching of behavioral sequences with the four-stage identification rules adopts a weighted scoring mechanism. Each stage sets 5-8 feature conditions, and each condition is assigned a different weight. When the sum of the weighted scores of all matching conditions exceeds the judgment threshold of that stage, the current behavioral sequence is identified as entering the crime script stage.

[0042] Specifically, the threshold for determining the preparation phase is 0.6, the threshold for the contact phase is 0.5, the threshold for the infringement phase is 0.7, and the threshold for the maintenance phase is 0.6. When the same session matches multiple phases simultaneously, the phase with the highest matching score is taken as the current phase.

[0043] Step S4: Risk Score Calculation. The system calculates a risk score for behavioral sessions that match the crime script stage.

[0044] Taking a specific case as an example: Suspect X (male, 25 years old) added 14-year-old Y as a friend through platform B and sent 127 messages in 3 days (the average interaction frequency of users of the same age on the platform is 15 messages / 3 days). The messages contained emotional manipulation language such as "Brother cares about you" and "This is our secret". On the 3rd day, he asked Y to send personal photos.

[0045] The calculations for each indicator are as follows: S1 = 127 / 15 = 8.47, after normalization, take 1.0 (it becomes saturated when it exceeds 5 times). S2: The content classification model scores 0.72 for sensitivity to interactive text; S3: According to the platform records, X received 2 reports. S3 = 2 × 0.2 = 0.4; S4: Y is 14 years old, S4 = 0.7; S5: X initiates interaction with Y on both platform B and platform C. The number of cross-platform interactions is 2, and S5 = 0.6.

[0046] R=0.20×1.0+0.30×0.72+0.15×0.4+0.20×0.7+0.15×0.6 =0.20+0.216+0.06+0.14+0.09=0.706 Since R ≥ 0.7, the system generates an emergency-level warning.

[0047] Then it is sent to the manual review platform for manual review. If the manual review determines that there is no risk or low risk, it is judged as a misjudgment, the warning is revoked, and monitoring needs to be stopped; if it is still judged as risky, then the subsequent operations continue.

[0048] Step S5: Tiered Intervention Implementation. The system automatically executes emergency-level intervention measures: freezes X's account, saves a snapshot of all conversation data between X and Y (a total of 127 messages, 3 image sending records, and 2 video call records), pushes case clues and evidence summaries to the connected law enforcement agency's information system, and simultaneously sends an alert notification (including a risk summary and recommended measures) to Y's registered guardian.

[0049] Step S6: Electronic Evidence Solidification. The system uses all interaction records between X and Y (including the original message text, user identity information, timestamp, and device information) as data to be preserved, calculates the SHA-256 hash digest value, and packages the digest value and the preservation timestamp into the consortium blockchain. The evidence chain contains 127 preserved transaction records, organized into a verifiable and complete evidence structure using a Merkle tree. Law enforcement agencies can verify any record through the verification interface.

[0050] Example 3

[0051] Cross-platform collaborative governance and victim protection.

[0052] This embodiment provides a cross-platform collaborative governance mechanism and a victim protection and assistance module in the governance system.

[0053] like Figure 3 As shown, the cross-platform collaborative governance module connects with the risk control systems of various Internet platforms through standardized interfaces to establish a unified user risk profile sharing mechanism.

[0054] Specifically, the cross-platform collaborative governance module maintains a global list of high-risk users, which stores the high-risk user identifiers and risk characteristic summaries reported by each platform. When a platform's (e.g., Platform A) local risk control system identifies a user as high-risk, Platform A reports the user's risk identifier (a unique anonymized identifier generated by the system), risk level, matched crime script stage, and key behavioral characteristic summary (excluding original privacy data) to the cross-platform collaborative governance module. After reviewing the reported information, this module pushes the risk identifier to the risk control systems of other connected platforms (Platforms B, C, D, and E). Upon receiving the push, each platform increases the monitoring level or implements preventative restrictions on the user corresponding to the risk identifier in its local system.

[0055] To prevent the misuse and misjudgment of risk information, the cross-platform collaborative governance module has set up the following security mechanisms: the push of risk labels only includes anonymized identifiers and risk levels, and does not include users' personal identity information or interactive content; each platform's response to the push information is subject to permission constraints, and is only allowed to upgrade the monitoring level and implement limited restrictions, and is not allowed to directly ban accounts (banning requires confirmation through a local review process); risk labels have an expiration period (30 days for observation level, 60 days for warning level, and permanent for emergency level), and will be automatically downgraded or cleared after expiration.

[0056] The victim protection and assistance module is automatically activated after a case is confirmed by law enforcement agencies, and the following protection procedures are executed: (1) Comprehensive investigation and deletion of privacy information across the entire network: The system extracts the hash value of privacy content from the solidified electronic evidence and performs hash matching retrieval on various connected platforms. Upon discovering relevant privacy content, the system automatically triggers a deletion command. For modified or cropped derivative versions, the system uses the Perceptual Hash Algorithm (pHash) for similarity matching, with a similarity threshold set at 0.85. Derivative content exceeding the threshold is also deleted. The comprehensive investigation cycle is as follows: the first investigation is performed immediately after the case is confirmed, followed by a continuous investigation every 7 days for a period of 6 months.

[0057] (2) Psychological assistance referral: The system automatically generates a psychological assistance referral form for victims. The referral form includes the victim's basic information, case summary, preliminary psychological assessment results and a list of recommended counseling institutions. It is pushed to the connected psychological counseling institution or juvenile protection organization through an encrypted channel.

[0058] (3) Protection File Management: The system establishes an electronic protection file for each victim, recording all rescue measures implemented, progress of psychological counseling, and family follow-up visits. The file has strict access control, and only authorized personnel can access it. All access operations are logged. The system automatically generates a follow-up reminder every 30 days, prompting staff to conduct regular follow-ups and status assessments of the victims.

[0059] Example 4

[0060] This embodiment provides a specific scenario of the governance system being deployed and applied in a provincial-level procuratorate.

[0061] A provincial-level procuratorate has deployed this governance system and established data connections with three major internet platforms (instant messaging platforms, social networking platforms, and online gaming platforms) within its jurisdiction.

[0062] The deployment process is as follows: In the initial stage of system launch, a full identity verification was conducted on all existing users who had joined the platform. A total of 12,356 suspected underage accounts were identified through age profile models. After manual review, 9,872 of these accounts were confirmed to be underage accounts, and the system automatically marked them with protection tags.

[0063] During daily operation, the system monitors the interactive behavior of underage users in real time. One day, the criminal script behavior analysis engine detected on platform A that user M (who was verified as an adult) sent a large number of messages to 5 underage users (aged 11-15) within 3 days. The message content analysis showed that M used enticing language such as "I am your fan club administrator" and "Send me photos to get autographed photos".

[0064] The behavior sequence matching results from the crime script behavior analysis engine show: User M's behavior was matched to the preparation stage (adding underage friends in bulk, using a fake identity) and the contact stage (sending high-frequency messages, using emotionally manipulative language), with matching confidence levels of 0.82 and 0.76, respectively.

[0065] Risk assessment module calculation results: S1=1.0 (interaction frequency is more than 8 times the average of peers), S2=0.65 (content sensitivity is moderate to high), S3=0.6 (M had a previous violation record on platform B), S4=0.7 (3 of the 5 victims are under 14 years old), S5=0.3 (related behavior was only detected on platform A).

[0066] R=0.20×1.0+0.30×0.65+0.15×0.6+0.20×0.7+0.15×0.3 =0.20+0.195+0.09+0.14+0.045=0.67 Since 0.4 ≤ R < 0.7, the system generates a warning level alert.

[0067] The intervention execution module automatically performs the following measures: limits M's message sending frequency to 3 messages per hour, prohibits adding new friends, sends risk warning notifications (including protection suggestions) to 5 underage victims and their guardians, and includes M's behavioral data in the key monitoring list.

[0068] After receiving a system-triggered warning on their enforcement terminal, prosecutors manually reviewed M's behavior. The review confirmed that M's actions exhibited clear signs of attempted indecent assault from a distance, prompting a decision to initiate further investigation. The system then used an electronic evidence storage module to write all of M's interaction data, identity information, and behavioral timeline into a blockchain for archiving.

[0069] In the subsequent investigation, although M's account was restricted, there were still signs of activity on other platforms. The cross-platform collaborative governance module pushed M's risk marker to platforms B and C, and both platforms immediately raised the monitoring level of M's account. A week later, platform B detected that M was attempting to continue contacting underage users under a new identity, and the system automatically triggered an emergency alert, freezing M's account on platform B. Based on the complete chain of evidence pushed by the system (covering the interaction records of platforms A and B), the procuratorate initiated an investigation against M in accordance with the law. During the case handling process, the blockchain-stored electronic evidence underwent integrity verification through a judicial verification interface and was accepted by the court as valid evidence.

[0070] After the case was confirmed, the victim protection and assistance module was automatically activated, generating psychological assistance referral forms for each of the five victims, initiating a full-network privacy information screening (the initial screening identified and deleted three related images), and establishing protection files for continuous follow-up visits.

[0071] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A governance system for preventing molestation of minors, characterized in that, include: The multi-source data acquisition module collects user interaction data in real time from at least two or more Internet platforms. The interaction data includes text messages, image information, voice information, video information, user registration information, login behavior data, and social relationship chain data. The identity verification and minor identification module performs dual verification of real-name authentication and face liveness detection for registered users. Based on user registration information, behavioral feature data and device fingerprint information, it builds a user age profile model to determine whether the user is a minor and marks the accounts of identified minor users with protection tags. The crime script behavior analysis engine constructs a behavior sequence model based on crime script theory, which includes a preparation phase, contact phase, infringement phase, and maintenance phase. It extracts temporal behavior features from the user interaction data, matches the extracted behavior features with the behavior sequence model, and identifies the crime script phase in which the current interaction behavior is located. The risk assessment and early warning module calculates a risk score based on the stage matching results output by the crime script behavior analysis engine and a preset risk assessment index system. When the risk score exceeds a preset threshold, it generates an early warning signal of the corresponding level based on the current stage of the crime script and pushes it to the intervention execution module. The intervention execution module performs graded intervention measures according to the level of the warning signal, including pushing risk warnings to underage users, restricting the behavior of suspected accounts, sending alarm notifications to guardians, and pushing case clues to law enforcement agencies; The electronic evidence solidification module automatically calculates hash values ​​for relevant interaction data, user identity information, and behavior timelines when it detects interaction behaviors that match the characteristics of a criminal script. The resulting hash values ​​are then written into the blockchain for storage, forming an electronic evidence chain with timestamps and tamper-proof attributes.

2. The governance system for preventing molestation of minors according to claim 1, characterized in that, The user age profile model in the identity verification and minor identification module adopts a multilayer perceptron network structure. The input layer receives user registration time, active period distribution, language style feature vector, and social relationship graph features. The output layer outputs the probability value that the user is a minor. When the probability value is greater than the preset age determination threshold, the user is marked as a suspected minor and a further identity verification process is triggered.

3. The governance system for preventing molestation of minors according to claim 1, characterized in that, The behavioral sequence model sets identification rules for each stage: the preparation stage detects whether the account is newly registered, frequently switches devices, or uses a virtual identity; the contact stage detects whether adult users initiate excessive message interactions with underage users or use emotionally manipulative language; the infringement stage detects whether the interaction content contains features such as soliciting private pictures or videos, threats, or intimidation; and the maintenance stage detects whether there is a behavioral pattern of continuously blackmailing or spreading private materials across platforms.

4. The governance system for preventing molestation of minors according to claim 1, characterized in that, The risk assessment index system includes five indicators: abnormality of interaction frequency, sensitivity of content, historical risk record of the actor, vulnerability of the victim's age, and correlation of cross-platform behavior. The risk score R = Σ(wi×Si), where Si is the normalized score value of the i-th item, wi is the weight, and the sum of the weights is 1. When R is less than the first threshold, it is an observation-level warning; when it is between the first and second thresholds, it is an alert-level warning; and when it is greater than or equal to the second threshold, it is an emergency-level warning.

5. The governance system for preventing molestation of minors according to claim 1, characterized in that, When writing to the blockchain, the electronic evidence solidification module takes the original content of the interaction data, the user identifiers of the sender and receiver, the message timestamp, and the device information as the data to be stored. It uses the SHA-256 algorithm to calculate the hash digest value of the data to be stored, packages the hash digest value together with the storage timestamp into a storage transaction record and writes it into the blockchain network. It then organizes multiple storage records through a Merkle tree structure to form a traceable and verifiable evidence chain.

6. The governance system for preventing molestation of minors according to claim 1, characterized in that, The intervention execution module implements tiered intervention measures according to the increasing warning level: when an observation-level warning is triggered, a safety risk education prompt is pushed to the underage user's terminal and the granularity of log recording is increased; when a warning-level warning is triggered, the message sending frequency and friend adding function of the suspected account are restricted and an alarm notification is sent to the guardian. When an emergency alert is triggered, the suspected account will be immediately frozen and all session data will be saved. Case leads and solidified electronic evidence will be simultaneously pushed to the law enforcement agency's information system.

7. The governance system for preventing molestation of minors according to claim 1, characterized in that, It also includes a cross-platform collaborative governance module, which connects with the risk control systems of multiple Internet platforms through standardized interfaces to establish a cross-platform user risk profile sharing mechanism. When a platform identifies a high-risk user, it synchronizes the risk identifier and feature summary of the high-risk user to the cross-platform collaborative governance module, which then pushes the risk identifier to other connected platforms to achieve cross-platform collaborative risk prevention and control.

8. The governance system for preventing molestation of minors according to claim 1, characterized in that, It also includes a victim protection and assistance module, which automatically initiates the victim protection process after the case is confirmed. The protection process includes: triggering a full network screening and deletion command for privacy information, searching and deleting relevant privacy content on various docking platforms through hash value matching; generating a victim psychological assistance referral form and pushing it to the docking psychological counseling institution; and establishing a victim protection file to record the implementation of assistance measures and regular follow-up information.

9. A method for preventing molestation of minors, applied to the governance system as described in any one of claims 1 to 8, characterized in that, Includes the following steps: S1. Collect user interaction data in real time from at least two Internet platforms, including text messages, image information, voice information, video information, user registration information, login behavior data, and social relationship chain data; S2. Perform dual verification of real-name authentication and face liveness detection on registered users. Based on user registration information, behavioral characteristics and device fingerprints, build a user age profile model to determine whether the user is a minor and mark the account of underage users with a protection label. S3. Based on the crime script theory, construct a behavioral sequence model that includes four stages: preparation, contact, infringement and maintenance. Extract temporal behavioral features from user interaction data, match the extracted behavioral features with the identification rules of each stage, and identify the crime script stage in which the current interaction behavior is located. S4. Based on the stage matching results, calculate the risk score R=Σ(wi×Si) using an indicator system that includes anomaly of interaction frequency, sensitivity of content, historical risk record of the actor, vulnerability of the victim's age, and cross-platform behavioral correlation. S5. When the risk score exceeds the preset threshold, a corresponding level warning signal is generated based on the current stage of the crime script, and graded intervention measures are implemented according to the level, including risk warning push, restriction of suspected accounts, warning to guardians, and push of case clues to law enforcement agencies. S6. When an interaction behavior that matches the characteristics of a criminal script is detected, the hash value of the interaction data, user identity information and behavior time chain is automatically calculated, and the hash value is written into the blockchain for storage, forming an immutable electronic evidence chain.

Images