A short message signature information intelligent authentication method and system

Abstract

The present application relates to the technical field of information communication, and in particular to a short message signature information intelligent authentication method and system. The technical problem of insufficient overall identification accuracy of abnormal signature information by the prior art is solved. The method comprises: obtaining historical authentication data of a plurality of short message signature subjects; clustering the plurality of short message signature subjects based on the historical authentication data, and determining at least one risk category; training a signature authentication model corresponding to each risk category based on the historical authentication data of the short message signature subjects in each risk category; and authenticating based on the risk category to which the sender of the short message information to be authenticated belongs and the corresponding signature authentication model. The present application is used in the short message signature authentication scene.

Description

Technical Field

[0001] This invention relates to the field of information and communication technology, specifically to a method and system for intelligent authentication of SMS signature information. Background Technology

[0002] SMS signatures, as a key element identifying the sender's identity, are widely used in various business scenarios such as corporate notifications and financial services. They serve as a primary basis for users to judge the credibility and compliance of information. Therefore, accurate authentication of SMS signature information has become a crucial technical step in ensuring communication security and preventing fraud and spam. Currently, common SMS signature authentication methods mainly rely on static comparison of pre-submitted registration information by the sender, or combine some simple sending frequency and content rules for risk screening. However, with the increasing number of SMS senders and the growing complexity of their behavioral patterns, existing methods are struggling to effectively address evolving evasion techniques, resulting in insufficient overall accuracy in identifying abnormal signature information. Summary of the Invention

[0003] To address the technical problem that existing methods are insufficient in effectively dealing with constantly evolving circumvention techniques, resulting in inadequate overall accuracy in identifying abnormal signature information, the present invention aims to provide an intelligent authentication method and system for SMS signature information. The specific technical solution adopted is as follows:

[0004] In a first aspect, the present invention provides an intelligent authentication method for SMS signature information. The method includes: acquiring historical authentication data of multiple SMS signature entities; the historical authentication data includes: registration information and historical sending records; the registration information is the identity and qualification information pre-registered by the SMS signature entity on a regulatory platform; the historical sending records are the historical SMS sending process and result data of the SMS signature entity; based on the historical authentication data, clustering the multiple SMS signature entities to determine at least one risk category; based on the historical authentication data of the SMS signature entities within each risk category, training a signature authentication model corresponding to each risk category; and based on the risk category to which the sender of the SMS message to be authenticated belongs, calling the corresponding signature authentication model for authentication.

[0005] In conjunction with the first aspect mentioned above, in one possible implementation, the behavioral characteristics include: a level characteristic of the number of messages sent, a frequency characteristic of anomalies, and a pattern characteristic of the sending time. The level characteristic of the number of messages sent is used to characterize the scale of the number of messages sent. The frequency characteristic of anomalies is used to characterize the historical ratio of messages sent that were judged as abnormal. The pattern characteristic of the sending time is used to characterize the distribution pattern of messages sent over time.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: obtaining a sequence of the number of messages sent by each SMS signature subject on multiple statistical dates based on the historical sending records of the SMS signature subject; the number of messages sent on the statistical dates, the number of abnormal messages, and the number of complaints; determining the level characteristics of the number of messages sent based on the fluctuation of the number of messages sent between multiple statistical dates; determining the frequency characteristics of abnormal occurrence based on the type distribution of abnormal SMS messages in the historical sending records; and determining the time pattern characteristics of sending based on the distribution of SMS sending times in the historical sending records.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: calculating the behavioral feature difference between every two SMS signing entities among multiple SMS signing entities; and based on the behavioral feature difference, using a preset clustering algorithm to cluster the multiple SMS signing entities to determine at least one risk category.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the reporting information includes: SMS signature text. The method specifically includes: determining a first similarity between each pair of SMS signature subjects based on their SMS signature texts; the first similarity characterizing the degree of similarity between the registered signature texts of the two SMS signature subjects; determining a second similarity between each pair of SMS signature subjects based on their behavioral characteristics; the second similarity characterizing the overall similarity of the historical sending behaviors of the two SMS signature subjects; and determining the behavioral characteristic difference degree based on the correlation between the first and second similarities.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: constructing a training sample set for the target risk category based on the SMS content in the historical sending records of all SMS signature subjects within the target risk category; the target risk category is any risk category among at least one risk category; training a preset text classification model based on the training sample set to determine the signature authentication model for the target risk category; and determining the signature authentication model for each risk category within at least one risk category.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, the method further includes: performing control operations matching the authentication result type based on the authentication result of the SMS message to be authenticated; the authentication result includes: compliance, violation, and suspected violation; the control operations include: when the authentication result is compliant, allowing the sending of the SMS message to be authenticated; when the authentication result is violation, intercepting the SMS message to be authenticated and generating a violation handling report; when the authentication result is suspected violation, transferring the SMS message to be authenticated to the manual review process.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the reporting information includes the enterprise's unified social credit code, SMS signature text, and reporting validity period; historical sending records include: SMS content body, sending time, interception status indicator, and complaint status indicator.

[0012] Secondly, this invention provides an intelligent authentication system for SMS signature information. The system includes: a data acquisition module for acquiring historical authentication data of multiple SMS signature entities; the historical authentication data includes: registration information and historical sending records; the registration information is the identity and qualification information pre-registered by the SMS signature entity on a regulatory platform; the historical sending records are the historical SMS sending process and result data of the SMS signature entity; a clustering analysis module for clustering multiple SMS signature entities based on the historical authentication data to determine at least one risk category; a model training module for training a signature authentication model corresponding to each risk category based on the historical authentication data of the SMS signature entities within each risk category; and an authentication execution module for calling the corresponding signature authentication model to perform authentication based on the risk category to which the sender of the SMS message to be authenticated belongs.

[0013] The present invention has the following beneficial effects:

[0014] This invention fundamentally changes the traditional, static verification model of SMS signature authentication by constructing a complete process for acquiring historical data, clustering risk categories, training models for each category, and invoking the model for authentication based on risk. It intelligently clusters mixed SMS senders based on their historical behavioral data, thereby identifying groups with similar inherent risk characteristics and training a dedicated authentication model for each group. This method achieves precision and dynamism in the authentication strategy, adapting to the behavioral differences of subjects in different risk categories. Therefore, it significantly improves the overall accuracy and risk resistance of the authentication system when facing constantly evolving evasion methods such as spam and fraudulent SMS messages. This solves the technical problem that existing methods struggle to effectively cope with constantly evolving evasion methods, leading to insufficient overall accuracy in identifying abnormal signature information. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a method for intelligent authentication of SMS signature information according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of an intelligent authentication system architecture for SMS signature information provided in one embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a smart authentication method and system for SMS signature information proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent authentication method and system for SMS signature information provided by this invention.

[0021] Please see Figure 1 The diagram illustrates a flowchart of an intelligent authentication method for SMS signature information according to an embodiment of the present invention. The method includes the following steps S101-S105, which will be described in detail below.

[0022] S101. Obtain historical authentication data of multiple SMS signature subjects.

[0023] The historical authentication data includes: registration information and historical sending records; the registration information is the identity and qualification information of the SMS signature subject pre-registered on the regulatory platform; the historical sending records are the historical SMS sending process and result data of the SMS signature subject.

[0024] One possible implementation involves establishing a stable connection with the SMS business systems and entity registration management platforms of the three major telecom operators through a standardized Application Programming Interface (API). The system then uses the Hypertext Transfer Protocol (HTTP) to read the registration information of each SMS signature entity from a pre-built synchronous database. This registration information includes, but is not limited to, the enterprise's unified social credit code, personal real-name information, registration signature style, business scope, and registration validity period. Simultaneously, the system obtains the complete content of each SMS signature entity's historical authenticated SMS messages, including but not limited to phone numbers, URLs, timestamps of sending, interception records, complaint records, and recipient location information.

[0025] S102. Based on historical authentication data, cluster multiple SMS signature entities to determine at least one risk category.

[0026] One possible implementation involves analyzing the historical authentication data of all SMS signature subjects to calculate a metric that reflects the comprehensive differences in behavioral patterns and identity between any two subjects. Based on this metric, all subjects are automatically grouped to identify groups with inherently similar risk characteristics.

[0027] For example, the system first extracts behavioral characteristics of each entity based on its historical sending records, considering multiple dimensions such as the number of sending transactions, the frequency of abnormal occurrences, and sending time preferences. By comprehensively comparing the similarity of the registered signature texts between any two entities, as well as the overall similarity of their multi-dimensional behavioral characteristics, a quantitative value representing the comprehensive difference between the two entities is calculated. Finally, based on the comprehensive difference between all pairs of entities, a preset clustering algorithm is used to automatically divide them, aggregating entities with similar behavioral patterns and signature usage habits into the same cluster. Each cluster is defined as a risk category with common characteristics.

[0028] S103. Based on the historical authentication data of SMS signature subjects within at least one risk category, train the signature authentication model corresponding to each risk category.

[0029] One possible implementation involves first extracting the text content of all historical SMS messages from the historical sending records of each entity belonging to any risk category, and labeling these messages based on whether they are related to violations, suspected violations, or compliance results, thereby constructing a training sample set for that risk category. Subsequently, this training sample set is used to train a pre-defined neural network classification model to determine the signature authentication model corresponding to that risk category; the above steps are repeated to determine the signature authentication model corresponding to each risk category.

[0030] S104. Based on the risk category of the sender of the SMS message to be authenticated, call the corresponding signature authentication model to perform authentication.

[0031] In one possible implementation, when a text message to be authenticated is received, the sender's identifier is first identified and extracted to determine the specific risk category to which the sender has been classified in historical data analysis. Then, based on the risk category identifier, a signature authentication model built for that risk category is invoked from a pre-trained model set. The complete content of the text message to be authenticated, its signature information, and relevant context are submitted as input to the invoked signature authentication model. Based on its deep understanding and feature recognition capabilities of text message content belonging to its risk category, the model performs real-time analysis and reasoning on the input information, ultimately outputting a clear authentication conclusion.

[0032] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment fundamentally changes the single, static verification mode in traditional SMS signature authentication by constructing a complete method and process of acquiring historical data, clustering and classifying risk categories, training models by category, and calling the model for authentication according to risk. It intelligently clusters mixed SMS senders based on their historical behavior data, thereby identifying groups with similar inherent risk characteristics and training a dedicated authentication model for each group. This method achieves precision and dynamism in the authentication strategy, adapting to the differences in behavioral patterns of subjects in different risk categories. Therefore, when facing constantly changing evasion methods such as spam and fraudulent SMS messages, it significantly improves the overall accuracy and risk resistance of the authentication system, thus solving the technical problem that existing methods are unable to effectively cope with constantly evolving evasion methods, resulting in insufficient overall accuracy in identifying abnormal signature information.

[0033] In one possible implementation, the process of clustering multiple SMS signature subjects based on historical authentication data to determine at least one risk category can be specifically implemented through the following S201-S202, which will be explained in detail below.

[0034] S201. Based on historical authentication data, determine the behavioral characteristics of each SMS signature subject.

[0035] Among them, behavioral characteristics are used to characterize the historical sending behavior patterns and risk status of SMS signature subjects.

[0036] One possible implementation involves conducting multi-dimensional quantitative analysis of the historical sending records of each SMS signature subject, extracting and constructing a set of quantitative indicators that can comprehensively characterize the subject's historical sending activity patterns and risk status, i.e., behavioral characteristics.

[0037] For example, firstly, based on the entity's historical sending records, a sequence of sending quantities over multiple consecutive statistical dates is constructed. This sequence includes the total number of SMS messages sent each day, the number blocked by the system due to anomalies, and the number of user complaints. Then, by analyzing the temporal fluctuations of this sending quantity sequence, a sending quantity level characteristic representing the stability of the entity's sending scale is determined. Next, by statistically analyzing the distribution of various violation types among the SMS messages marked as abnormal in its historical sending records, anomaly frequency characteristics representing the entity's historical violation behavior preferences are determined. Finally, by analyzing the distribution patterns of all historical SMS sending times, sending time pattern characteristics representing the entity's operational habits over time are determined.

[0038] S202. Based on the behavioral characteristics of multiple SMS signing entities, cluster the multiple SMS signing entities.

[0039] In one possible implementation, after obtaining the behavioral characteristics of all SMS signature subjects, the comprehensive difference between any two subjects is quantitatively calculated, and all subjects are grouped according to this difference, thereby aggregating subjects with similar behavioral patterns into the same category.

[0040] For example, the behavioral feature difference between any two SMS signature subjects is first calculated. This difference calculation comprehensively considers the overall similarity between the two in terms of historical sending behavior features and the similarity between their registered signature texts. This comprehensive difference is used as the core metric for measuring the similarity between subjects. A preset clustering algorithm is used to divide all SMS signature subjects, and subjects with small differences between each other are aggregated into the same group. Finally, at least one risk category set with strong internal homogeneity and obvious heterogeneity between groups is output.

[0041] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment further defines the clustering process by including two core steps: determining behavioral characteristics and clustering based on behavioral characteristics, thus providing a data analysis basis for risk category classification. Its beneficial effect lies in concretizing the abstract concept of risk into quantifiable and comparable behavioral characteristics, making clustering analysis no longer a simple data grouping, but rather a deep dive into the historical sending behavior patterns and risk status of the subject. This ensures that the final risk categories can truly reflect the inherent behavioral commonalities and risk correlations within the group, providing a reliable premise for subsequently building a high-precision, highly targeted classification and authentication model, and avoiding the instability of results caused by clustering based on superficial or random features.

[0042] In one possible implementation, the behavioral features include: sending quantity level features, anomaly frequency features, and sending time pattern features; the sending quantity level features include: sending magnitude sub-features and sending stability sub-features, the sending magnitude sub-features are used to characterize the quantity scale of SMS sending, and the sending stability sub-features are used to characterize the degree of fluctuation in SMS sending behavior; the anomaly frequency features are used to characterize the historical ratio of SMS sending being judged as abnormal; and the sending time pattern features are used to characterize the distribution pattern of SMS sending behavior in the time dimension.

[0043] The technical solution provided by the above embodiments can bring at least the following beneficial effects: by clearly defining behavioral characteristics as three dimensions—the level of sending quantity, the frequency of anomalies, and the pattern of sending time—a clear and comprehensive quantitative framework is provided for behavioral analysis. Its beneficial effect lies in comprehensively characterizing the three dimensions that best reflect the behavioral patterns and potential risks of the sending entity: business scale (quantity level), compliance status (anomaly frequency), and operational habits (time pattern). This allows for a three-dimensional construction of the entity's behavioral profile, greatly enriching the information available for risk assessment and improving the representativeness and discriminative power of behavioral characteristics.

[0044] In one possible implementation, the process of determining the behavioral characteristics of each SMS signature subject based on historical authentication data can be specifically implemented through the following S301-S304, which will be explained in detail below.

[0045] S301. Based on the historical sending records of the SMS signature subject, obtain the sending quantity sequence of each SMS signature subject on multiple statistical dates.

[0046] The number of messages sent includes: the number of messages sent, the number of exceptions, and the number of complaints on the statistical date.

[0047] In one possible implementation, for each SMS signature subject, periodic statistics are performed based on its detailed historical sending records to construct a sequence data reflecting the core quantitative indicators of the subject's sending activities in different statistical periods.

[0048] For example, using a calendar day as a statistical period, each sending record of the entity within a preset historical time period (such as the past three months) is traversed. For each statistical date, the total number of SMS messages successfully sent by the entity on that date is calculated to determine the sending quantity; the number of SMS messages automatically blocked due to their sending behavior triggering preset exception rules on that date is counted to determine the exception quantity; simultaneously, the number of valid complaints filed by recipient users against SMS messages sent by the entity on that date is counted to determine the complaint quantity. Arranging these three quantities for each statistical date in chronological order constitutes the sending quantity sequence of the entity.

[0049] Understandably, this step is a crucial preprocessing stage that transforms discrete, event-level historical behavior logs into continuous, structured statistical data suitable for time-series analysis and pattern recognition. By constructing a sequence of sent data, the system abstracts the raw behavioral flow data into standardized indicators that comprehensively reflect the daily activity status of SMS signature subjects from three key dimensions: business volume, compliance, and user feedback. This sequence data provides the most direct and fundamental quantitative input for subsequent analysis of the stability of subject behavior, identification of abnormal fluctuation patterns, and calculation of differences in behavioral characteristics across dates.

[0050] S302. Based on the fluctuation of the transmission quantity sequence across multiple statistical dates, determine the level characteristics of the transmission quantity.

[0051] In one possible implementation, time-series analysis is performed on the transmission quantity sequence. By quantitatively evaluating the transmission quantity sequence, a transmission quantity sub-feature is determined. This feature is used to quantify the business scale of the subject and can be determined by directly calculating the statistical value of the transmission quantity component in the transmission quantity sequence.

[0052] For example, the average and median of the number of messages sent over all statistical dates can be calculated as characteristics of the sending volume. The larger this value, the larger the scale of the SMS sending business of the entity.

[0053] In one possible implementation, after obtaining the sending quantity sequence, to further analyze the stability and risk fluctuation patterns of the SMS signature subject's behavior, a time-series difference analysis can be performed on the sequence. The daily data (sending volume, abnormal volume, complaint volume) in the sending quantity sequence is treated as a multi-dimensional feature point, and the degree of difference between the feature points corresponding to different statistical dates is calculated (e.g., calculating Euclidean distance). This difference measure quantifies the degree of deviation of the subject's daily sending behavior characteristics; the greater the difference, the more unstable the behavior and the higher the probability of abnormal sending cycles. Simultaneously, a time-series comparison of the distribution of historical violations and complaint types can be performed. By calculating the probability difference of the type distribution between different dates (e.g., KL divergence), the evolution of its violation patterns can be assessed. Finally, based on the comprehensive analysis of such time-series differences, a sending stability sub-feature is determined to characterize the fluctuation level of the subject's SMS sending behavior. For example, the first... day and day Inter-day transmission stability sub-features Satisfy the following formula 1:

[0054]

[0055] in, Indicates the first day and day Transmission stability sub-feature between days; Indicates the first day and day Euclidean distance between the feature sequences of a day; Indicates the first day and day The difference in probability statistics of violation types and complaint types of SMS messages sent between days; the larger the calculated feature value, the greater the difference in normal characteristics of the current SMS signature subject in the daily sending process; norm is a standardization function used to eliminate the interference of different parameters in magnitude and dimension, and the calculation results of different subjects are mapped to a unified comparable scale by the Z-score standardization method based on global statistics. and The preset weighting coefficients, + =1, for example, =0.5, =0.5.

[0056] For example, difference value The distance is obtained by calculating the statistical distance between the probability distribution vectors of violation and complaint types on day x and day y. In a preferred embodiment, the Jensen-Shannon Divergence (JS divergence) is used for measurement. JS divergence is a symmetric, bounded measure of the difference between two probability distributions; the larger the value, the greater the difference in the violation and complaint patterns between the two days. It is understood that the specific method for calculating b_(x,y) is not limited to JS divergence. Those skilled in the art can also use other well-known methods for measuring the difference in probability distributions, such as Hellinger distance, Bach distance, or directly calculate the Euclidean distance or cosine distance between the distribution vectors, as appropriate. This invention does not limit this. S303. Based on the type distribution of abnormal SMS messages in historical sending records, determine the frequency characteristics of abnormal occurrences.

[0057] One possible implementation involves classifying and statistically analyzing SMS messages marked as abnormal in the historical sending records of the SMS signature subject. By quantitatively assessing the composition ratio and concentration trend of each category of abnormality in its overall abnormal history, a comprehensive indicator is determined to characterize the specific risk preferences and high-incidence issues exhibited by the subject at the compliance level.

[0058] For example, based on a pre-defined violation classification system (e.g., unregistered signature, suspected fraudulent content, click fraud, spam marketing, etc.), each abnormal SMS message is automatically categorized and counted. Subsequently, the distribution structure of these abnormal types is analyzed: the proportion of each violation type in the total number of abnormalities is calculated, forming a type probability distribution. Based on the probability statistics of this type distribution and its changing patterns over time, the system determines a comprehensive indicator to characterize the specific risk preferences and concentration trends exhibited by the entity in historical abnormal activities, namely, the frequency of abnormal occurrence. For example, this characteristic can reflect whether the entity is mainly involved in high-risk violations such as fraud, or concentrated on marketing violations, and whether its violation strategies are singular or diverse.

[0059] S304. Based on the distribution of SMS sending times in historical sending records, determine the regularity of sending time patterns.

[0060] One possible implementation involves extracting the precise timestamps of all historical SMS messages sent by the subject and converting them to hours or finer-grained time units in a 24-hour format. By statistically analyzing the distribution of SMS volume across different time periods, the subject's peak sending periods (i.e., periods with significantly higher-than-average sending volume) and inactive periods are identified. Further analysis of the regularity of these peak periods is conducted, such as whether they consistently occur during weekday daytime or frequently appear during abnormal service hours like the early morning. Additionally, the consistency of active periods across different dates can be calculated (e.g., by calculating the intersection-union ratio of period distributions) to determine the stability of their time preferences. Based on a comprehensive analysis of the distribution of sending times, the identification of peak sending periods, and their regularity and consistency, a quantitative indicator is determined to holistically characterize the subject's operational habits and preferences in the time dimension—the sending time regularity characteristic. This characteristic can effectively distinguish subjects with normal work-rest schedules from those with high-frequency activity during abnormal times; the latter are typically associated with automated scripts, malicious harassment, or fraudulent behavior.

[0061] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment ensures the objectivity and operability of the feature extraction process by specifically defining the determination method of three behavioral features. Its beneficial effects are that, based on temporal fluctuations, it determines the quantity level feature, keenly capturing abnormal changes in the subject's transmission stability; based on the distribution of abnormal types, it determines the frequency feature, identifying specific risk points of preference violations; and based on the distribution of transmission times, it determines the temporal regularity feature, effectively identifying suspicious patterns such as operations during irregular time periods. This specific transformation method from raw data to feature values ​​not only makes feature calculation based on evidence, but also ensures that the extracted features have clear business meaning, making the subsequent risk analysis results more interpretable and providing high-quality, understandable input data for model training.

[0062] In one possible implementation, the process of clustering multiple SMS signature subjects based on their behavioral characteristics can be specifically implemented through the following steps S401-S402, which will be explained in detail below.

[0063] S401. Calculate the behavioral characteristic difference between any two SMS signature subjects among multiple SMS signature subjects.

[0064] In one possible implementation, multiple SMS signature subjects are paired up in pairs. For each pair of subjects, a scalar value reflecting the degree of overall difference between them is determined by comprehensively calculating their quantitative similarity at both the identity identification and behavioral pattern levels.

[0065] For example, for any two SMS signature entities, the system first extracts their registered signature texts from the filing information. By analyzing the character composition, semantic content, or word sequence of the texts, a first similarity value representing the similarity of their signature texts is calculated. Simultaneously, combining the determined multi-dimensional behavioral characteristics of the two entities (i.e., sending quantity level characteristics, anomaly frequency characteristics, and sending time pattern characteristics), a second similarity value representing the similarity of their historical sending behavior patterns is calculated by comprehensively measuring the overall closeness of these feature vectors in the numerical space. Finally, the system maps the first and second similarities together into a final behavioral feature difference value based on a preset association rule between them.

[0066] S402. Based on the difference in behavioral characteristics, use a preset clustering algorithm to cluster multiple SMS signature subjects and determine at least one risk category.

[0067] In one possible implementation, the calculated set of behavioral feature differences, reflecting the overall degree of difference between all SMS signature subjects, is used as the core input data and fed into a pre-defined clustering algorithm to perform automated group partitioning. First, based on the behavioral feature differences among all subjects, a matrix representing the internal differences within the entire subject set is constructed. Then, a pre-defined clustering algorithm (e.g., a density-based clustering algorithm) is invoked. This algorithm automatically identifies and aggregates subsets of subjects with low differences (i.e., high similarity) based on the relative magnitude of the differences between subjects, while separating subjects with high differences. Each identified subset with high internal similarity and large external differences is defined as a risk category with common risk characteristics.

[0068] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment automates and intelligently classifies risk groups by defining the clustering process as calculating the difference in behavioral characteristics and using a preset clustering algorithm. Its beneficial effect lies in that, by quantifying the difference in behavioral characteristics between any two subjects, the similarity comparison between subjects is transformed into a computable metric problem, providing accurate input for subsequent clustering algorithms. Combined with the automatic classification capabilities of preset clustering algorithms (such as density clustering, hierarchical clustering, etc.), massive numbers of subjects can be efficiently and accurately classified into different risk clusters, avoiding the subjectivity and inefficiency of manual classification, ensuring the consistency and scalability of risk classification results, and adapting to scenarios with dynamically increasing numbers of subjects.

[0069] In one possible implementation, the reporting information includes: SMS signature text. The process of calculating the behavioral feature difference between every two SMS signature subjects can be specifically implemented through the following S501-S502, which will be explained in detail below.

[0070] S501. Based on the SMS signature text between every two SMS signature subjects among multiple SMS signature subjects, determine the first similarity between every two SMS signature subjects.

[0071] In one possible implementation, for each pair of SMS signature entities, the standardized signature texts registered with the regulatory platform are extracted. A pre-defined text similarity calculation method is used to quantitatively assess the similarity between the two texts. Necessary preprocessing is performed on the two signature texts, such as removing irrelevant characters, standardizing character encoding, or performing word segmentation, to standardize the text units to be compared. Subsequently, a pre-defined text similarity calculation model or algorithm is used to analyze the two processed signature texts. This calculation process aims to capture commonalities in character sequence composition, word usage, or semantics, ultimately outputting a quantitative numerical result to characterize the degree of similarity between the two at the identity identification level, i.e., the first similarity.

[0072] For example, the signature is first segmented and standardized, then the signature string is converted into a semantic vector using the Word2Vec model, and finally the cosine similarity between the two semantic vectors is calculated. On the other hand, the similarity of SMS signatures can also be calculated using the Levenstein distance. The greater the similarity, the closer the information tampering such as homophone substitution is in the historical authentication information of the two entities, the higher the possibility that they use similar avoidance methods in the SMS signature authentication process, and the greater the probability of associated violation risks.

[0073] S502. Determine the second similarity between each pair of SMS signature subjects based on the behavioral characteristics between each pair of SMS signature subjects among multiple SMS signature subjects.

[0074] In one possible implementation, the following steps are performed to determine a second similarity between two SMS signature subjects (denoted as Subject A and Subject B), which is used to quantify the degree of similarity between the two in their historical sending behavior patterns.

[0075] First, acquire the behavioral features, encompassing multiple dimensions, already determined for subject A and subject B. These dimensions include at least: the quantity of transmissions representing the level and stability of transmission scale; the frequency of anomalies representing historical violation type preferences; and the transmission time pattern representing time-based operational habits. Then, for each dimension of behavioral features, calculate the feature value similarity between subject A and subject B in that dimension. If a feature is represented as a numerical vector, the similarity score for that dimension can be obtained by calculating the cosine similarity between two vectors; if the feature is represented as a probability distribution, a similarity measure between distributions (such as the Bach coefficient) can be calculated. After calculating the individual similarities for all preset dimensions, these individual results are merged into a single, scalarized value using preset integration rules (e.g., weighted average or geometric average of the similarity scores for each dimension). This final generated value is the second similarity score.

[0076] For example, the similarity of the sending quantity level features The following formula 2 is satisfied:

[0077]

[0078] in, This indicates the similarity between two entities in terms of the quantity of data sent. Send a quantity-level feature vector to subject A; Send a quantity-level feature vector to subject B; for The Euclidean norm of a vector; for The Euclidean norm of a vector; For parameter tuning coefficients, if When the value is 0, it is set to the minimum value other than 0. Dimensions and same.

[0079] To measure the degree of collinearity between two vectors. Used for normalization. The closer the result value is to 1, the more consistent the directions of the two vectors are, that is, the more similar their behavioral patterns are in that dimension; the closer it is to 0, the less correlated they are; a negative value indicates opposite patterns.

[0080] For example, the similarity of anomaly occurrence frequency features The following formula 3 is satisfied:

[0081]

[0082] in, Similarity of anomaly occurrence frequency characteristics; This represents the percentage of violations of type i by subject A; This represents the percentage of violations of type i in subject B; n is the total number of violation types. The Bach coefficient is used to measure the degree of overlap between two discrete probability distributions. It measures the similarity of the distributions of two subjects’ preferences for historical violation types.

[0083] For example, similarity of sending time pattern features Satisfy the following formula 4:

[0084]

[0085] in, This indicates the similarity between two entities in terms of their transmission time patterns. Send time-regularity feature vectors to subject A; Send time-regularity feature vectors to subject B; for The Euclidean norm of a vector; for The Euclidean norm of a vector; For parameter tuning coefficients, if When the value is 0, it is set to the minimum value other than 0. Dimensions and same.

[0086] The similarity of the distribution patterns of historical sending behaviors between two SMS signature subjects over the time dimension was quantified.

[0087] Understandably, this implementation method, through a strategy of first comparing dimensions and then integrating them, organically merges the similarity assessments of different aspects of the subject's behavior into a holistic index. This second similarity score, along with the first similarity score obtained from the signature text, are independent yet complementary, together forming the quantitative basis for comprehensively assessing the differences and connections between the two subjects, thereby supporting subsequent clustering algorithms in making accurate risk category classifications.

[0088] S503. Based on the correlation between the first similarity and the second similarity, determine the behavioral feature difference degree.

[0089] In one specific implementation, after determining the first similarity, which represents the similarity of the signature text between two SMS signing entities, and the second similarity, which represents the overall similarity of their historical sending behavior patterns, the final behavioral feature difference is calculated according to a preset rule that reflects the inherent relationship between the two entities. The relationship aims to construct a comprehensive metric so that the behavioral feature difference can simultaneously and sensitively reflect the similarities and differences between the two entities at both the identity identification and behavioral pattern levels.

[0090] For example, behavioral characteristic difference The following formula 5 is satisfied:

[0091]

[0092] in, The degree of difference in behavioral characteristics between any two SMS signature subjects; The first similarity score, The second similarity; This represents an exponential function with the natural constant as its base, avoiding a denominator of 0; For example, a preset positive real amplification factor (e.g., = 1). It is a dimensionless hyperparameter used to control the first similarity. The amplification intensity of the final difference. The larger the value, the higher the first similarity ( In cases of large differences, behavioral differences (1- The greater the magnification factor, the higher the magnification factor. Second similarity Converting to behavioral variability is a simple linear transformation with a range of [0, 1]. It is a comprehensive behavioral difference measure that has been adjusted for signature similarity risk. It not only reflects the original differences in behavioral patterns between two subjects, but also emphasizes the high-risk combination of signature imitation accompanied by behavioral abnormalities through an index mechanism, so that such subjects can be significantly distinguished in the difference space.

[0093] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment creatively integrates the static qualification marks and dynamic behavioral patterns of the subject for comprehensive risk assessment by introducing and combining the first similarity (signature text similarity) and the second similarity (behavioral feature similarity) to calculate the final behavioral feature difference. Its beneficial effect is that it overcomes the randomness of overlooking the possibility of similar behavior but unrelated signatures when only comparing behaviors, and can more effectively identify new evasion methods that imitate the signatures of legitimate subjects (high first similarity) but exhibit abnormal behavior (low second similarity), or discover potential illegal groups with highly coordinated behavioral patterns (high second similarity), thus enhancing the system's ability to identify complex evasion strategies.

[0094] In one possible implementation, the process of training the signature authentication model corresponding to each risk category based on the historical authentication data of the SMS signature subject within each risk category can be implemented through the following S601-S603, which will be described in detail below.

[0095] S601. Based on the SMS content in the historical sending records of all SMS signature subjects within the target risk category, a training sample set for the target risk category is constructed.

[0096] One possible implementation involves building a dedicated model training data foundation for each risk category determined through clustering (i.e., the target risk category). First, the complete text of all historical SMS messages is extracted from the historical sending records of each SMS signature subject belonging to that target risk category. This SMS content constitutes the initial raw corpus.

[0097] Subsequently, a corresponding risk label is generated for each SMS message to form the samples required for supervised learning. The labels are generated based on the association data generated during the historical sending of the SMS message: if the SMS message was blocked by the system, it is labeled as a violation; if the SMS message triggered a user complaint, it is labeled as a suspected violation; if the SMS message was neither blocked nor triggered a complaint, it is labeled as compliant. After data extraction and labeling are completed, all labeled SMS messages are integrated to form the initial training sample set for the target risk category.

[0098] In one possible implementation, when constructing the training sample set, if the sample category distribution within the target risk category is unbalanced, data from the global compliance sample library is introduced as negative samples to supplement the training sample set, or the SMOTE algorithm is used to generate synthetic samples of opposing categories to ensure that the training set contains diverse data on compliance, violations, and suspected violations.

[0099] S602. Based on the training sample set, train the preset text classification model to determine the signature authentication model for the target risk category.

[0100] One possible implementation involves using a training sample set built for a specific target risk category to perform supervised machine learning training on a signature authentication model specific to that category. First, a pre-defined text classification neural network model (such as a Text CNN, RNN, or Transformer architecture) is selected as the base model architecture. The core of the training process is to enable the model to learn to identify risk patterns and compliance characteristics related to that target risk category from the text message content.

[0101] For example, for each type of SMS signature subject, the historical authentication information data of all SMS signature subjects are divided into training, testing, and verification sets in a 7:2:1 ratio. A text convolutional neural network model is used for SMS signature information authentication, with a weighted cross-entropy loss function. To address the imbalance problem among the three types of samples (violation, suspected violation, and compliance), violation samples are assigned 3 times the weight, and suspected violation samples are assigned 2 times the weight, enhancing the model's ability to identify high-risk samples. The optimizer is the AdamW optimizer, with an initial learning rate of 0.002. A cosine annealing learning rate scheduling strategy is used, with 10 epochs per cycle, 40 training rounds, and a batch size of 32. If there is no improvement for 5 consecutive epochs, an early stop mechanism is triggered. Based on the above training, a dedicated SMS signature authentication model for specific risk categories is obtained. By deeply fitting the signature semantic features of the subject, the abnormal risk of SMS sending, and the characteristics of illegal tampering, the model can effectively and accurately identify anomalies in SMS signature information, such as homophonic tampering in promotional signatures of e-commerce subjects and splitting of lead generation phone numbers by financial subjects, thereby reducing the false negative rate for similar risks.

[0102] S603. Determine the signature authentication model for each risk category within at least one risk category.

[0103] In one possible implementation, for each risk category obtained through clustering, the aforementioned sample set construction and model training process is iteratively executed. That is, the current risk category to be processed is taken as the new target risk category, and steps S601 and S602 are repeated: extracting and labeling the historical SMS content of all subjects within the category to construct its own training sample set; subsequently, the preset text classification model is trained using this sample set, and the optimized model that meets the performance requirements is saved.

[0104] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment constructs a training set by extracting the internal SMS content of each risk category and training a specialized text classification model, thus achieving a customized authentication model. Its beneficial effect lies in the fact that SMS content for different risk categories often has different semantic features, word usage habits, and violation patterns (such as financial fraud rhetoric being drastically different from e-commerce spam marketing rhetoric). The model trained using the category's own data can deeply learn and fit the characteristics of this type of content, thereby having higher sensitivity and accuracy in SMS authentication under this category, significantly reducing interference between different risk patterns, effectively improving the model's recognition accuracy for various specific risks, and reducing misjudgments across risk scenarios.

[0105] In one possible implementation, after authenticating the SMS message based on the risk category of the sender, the corresponding signature authentication model is invoked. Then, control operations need to be performed based on the authentication results. The authentication results include: compliance, violation, and suspected violation. The control operations include: when the authentication result is compliance, allowing the SMS message to be authenticated to be sent; when the authentication result is violation, intercepting the SMS message to be authenticated and generating a violation handling report; when the authentication result is suspected violation, transferring the SMS message to be authenticated to the manual review process.

[0106] The technical solution provided by the above embodiments can bring at least the following beneficial effects: by adding a step to perform hierarchical control operations based on authentication results, a complete control loop from risk identification to risk handling is achieved. Its beneficial effect lies in the fact that it can not only determine risks, but also trigger differentiated handling actions (release, manual review, interception and reporting) based on the risk level (compliant, suspected violation, violation). This hierarchical processing mechanism takes into account both communication efficiency and security supervision, ensuring that high-risk information is effectively blocked while providing a buffer channel for manual review of uncertain suspected violation text messages.

[0107] In one possible implementation, the reporting information includes the enterprise's unified social credit code, SMS signature text, and reporting validity period; historical sending records include: SMS content body, sending time, interception status indicator, and complaint status indicator.

[0108] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment, by clearly defining the specific data items included in the reporting information and historical transmission records, lays a clear and standardized data foundation for the entire method. Its beneficial effect lies in unifying and standardizing the format and content of data input, ensuring that the data relied upon by all subsequent analysis and processing steps is complete, valid, and structured.

[0109] Please see Figure 2This illustration shows a schematic diagram of an intelligent authentication system architecture for SMS signature information provided by an embodiment of the present invention. The system includes: a data acquisition module 201, used to acquire historical authentication data of multiple SMS signature entities; the historical authentication data includes: registration information and historical sending records; the registration information is the identity and qualification information of the SMS signature entity pre-registered on the regulatory platform; the historical sending records are the historical SMS sending process and result data of the SMS signature entity; a clustering analysis module 202, used to cluster multiple SMS signature entities based on the historical authentication data to determine at least one risk category; a model training module 203, used to train a signature authentication model corresponding to each risk category based on the historical authentication data of the SMS signature entities within each risk category; and an authentication execution module 204, used to call the corresponding signature authentication model to perform authentication based on the risk category to which the sender of the SMS message to be authenticated belongs.

[0110] The technical solutions provided in the above embodiments can bring at least the following beneficial effects: This embodiment clarifies the physical implementation architecture of the invention by providing a system embodiment corresponding to the method. Its beneficial effect lies in mapping the method flow into independent functional modules such as data acquisition, cluster analysis, model training, and authentication execution, making the system design hierarchy clear and responsibilities well-defined, which is conducive to software development, deployment, maintenance, and expansion.

[0111] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0112] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for intelligent authentication of SMS signature information, characterized in that, The method includes: Obtain historical authentication data of multiple SMS signature entities; the historical authentication data includes: registration information and historical sending records; the registration information is the identity and qualification information of the SMS signature entity pre-registered on the regulatory platform, including: SMS signature text; the historical sending records are the historical SMS sending process and result data of the SMS signature entity; Based on the historical authentication data, the behavioral characteristics of each SMS signature subject are determined; the behavioral characteristics are used to characterize the historical sending behavior patterns and risk status of the SMS signature subject. Based on the SMS signature text between every two SMS signature entities among the plurality of SMS signature entities, a first similarity is determined between every two SMS signature entities; the first similarity is used to characterize the degree of similarity between the registration signature text of the two SMS signature entities; A second similarity is determined between every two SMS signature subjects based on the behavioral characteristics between them; the second similarity is used to characterize the overall similarity of the historical sending behaviors of the two SMS signature subjects. Based on the correlation between the first similarity and the second similarity, the difference in behavioral features is determined; Based on the behavioral feature differences, a preset clustering algorithm is used to cluster the multiple SMS signature subjects to determine at least one risk category. Based on the historical authentication data of the SMS signature subject within each of the at least one risk category, train the signature authentication model corresponding to each risk category. Based on the risk category of the sender of the SMS message to be authenticated, the corresponding signature authentication model is invoked for authentication.

2. The method for intelligent authentication of SMS signature information according to claim 1, characterized in that, The behavioral characteristics include: the level of sending quantity characteristics, the frequency of anomalies characteristics, and the regularity of sending time characteristics; the level of sending quantity characteristics are used to characterize the scale of SMS sending; the frequency of anomalies characteristics are used to characterize the historical ratio of SMS sending being judged as abnormal; and the regularity of sending time characteristics are used to characterize the distribution pattern of SMS sending behavior in the time dimension.

3. The method for intelligent authentication of SMS signature information according to claim 2, characterized in that, The process of determining the behavioral characteristics of each SMS signature subject based on the historical authentication data includes: Based on the historical sending records of the SMS signature subject, obtain the sending quantity sequence of each SMS signature subject on multiple statistical dates; the sending quantity sequence includes: the sending quantity, the number of abnormal cases, and the number of complaints on the statistical dates; The level characteristics of the transmission quantity are determined based on the fluctuation of the transmission quantity sequence among the multiple statistical dates; Based on the distribution of abnormal SMS types in the historical sending records, the frequency characteristics of the abnormality are determined; Based on the distribution of SMS sending times in the historical sending records, the regularity of the sending time is determined.

4. The method for intelligent authentication of SMS signature information according to claim 1, characterized in that, The step of training a signature authentication model for each risk category based on historical authentication data of SMS signature subjects within each of the at least one risk category includes: The training sample set for the target risk category is constructed based on the SMS content in the historical sending records of all SMS signature subjects within the target risk category; the target risk category is any one of the at least one risk category. Based on the training sample set, a preset text classification model is trained to determine the signature authentication model for the target risk category; Determine the signature authentication model for each risk category within the at least one risk category.

5. The method for intelligent authentication of SMS signature information according to claim 1, characterized in that, The method further includes: Based on the authentication result of the SMS message to be authenticated, perform control operations that match the authentication result type; the authentication result includes: compliance, violation, and suspected violation; the control operations include: when the authentication result is compliance, allow the SMS message to be authenticated to be sent; when the authentication result is violation, intercept the SMS message to be authenticated and generate a violation handling report; when the authentication result is suspected violation, transfer the SMS message to be authenticated to the manual review process.

6. The method for intelligent authentication of SMS signature information according to claim 1, characterized in that, The reported information includes the enterprise's unified social credit code, SMS signature text, and the reporting validity period; The historical sending records include: SMS content body, sending time, interception status indicator, and complaint status indicator.

7. A smart authentication system for SMS signature information, characterized in that, include: The data acquisition module is used to acquire historical authentication data from multiple SMS signature entities; The historical authentication data includes: registration information and historical sending records; the registration information is the identity and qualification information of the SMS signature subject pre-registered on the regulatory platform, including: SMS signature text; the historical sending records are the historical SMS sending process and result data of the SMS signature subject; The clustering analysis module is used to determine the behavioral characteristics of each SMS signature subject based on the historical authentication data; the behavioral characteristics are used to characterize the historical sending behavior patterns and risk status of the SMS signature subject. Based on the SMS signature text between every two SMS signature entities among the plurality of SMS signature entities, a first similarity is determined between every two SMS signature entities; the first similarity is used to characterize the degree of similarity between the registration signature text of the two SMS signature entities; A second similarity is determined between every two SMS signature subjects based on the behavioral characteristics between them; the second similarity is used to characterize the overall similarity of the historical sending behaviors of the two SMS signature subjects. Based on the correlation between the first similarity and the second similarity, the difference in behavioral features is determined; Based on the behavioral feature differences, a preset clustering algorithm is used to cluster the multiple SMS signature subjects to determine at least one risk category. The model training module is used to train the signature authentication model corresponding to each risk category based on the historical authentication data of the SMS signature subject in each risk category within the at least one risk category. The authentication execution module is used to call the corresponding signature authentication model to perform authentication based on the risk category of the sender of the SMS message to be authenticated.

Images