Multi-segment Sigmoid chain digital content popularity modeling method based on security risk adjustment

Abstract

The invention discloses a digital content popularity modeling method on a multi-segment Sigmoid chain based on security risk adjustment, which identifies a key turning point by introducing a moving average difference method (MAD), and sets a threshold value based on a mean value and a standard deviation or a percentile to effectively screen out a real market change signal. And carrying out quantitative modeling by using a segmented Sigmoid function, respectively carrying out fitting on each screened interval, and optimizing parameters of each segment of Sigmoid function by using a trusted region reflection (TRR) algorithm. And finally, carrying out weighted summation on each section of Sigmoid function, so as to obtain a popularity quantification result of each color. In addition, a systematic security event risk analysis module is introduced for the first time to dynamically adjust an original modeling result. Therefore, the method has remarkable innovation and optimization in the aspects of data sources, processing methods, dynamic change capture and quantitative modeling, the defects in the prior art can be effectively overcome, and a comprehensive, accurate and dynamic popularity quantitative tool is provided.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of statistical analysis and machine learning, and particularly relates to a method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment. BACKGROUND

[0002] In the existing technical system for analyzing the popularity of digital content, there are mainly two major analysis perspectives, and these two perspectives are often independent of each other, lacking effective integration and comprehensive consideration.

[0003] According to the existing media data and on-chain transaction data, there is a lack of effective integration mechanism. This fragmentation leads to the inability to fully reflect the real performance of digital content in the market when evaluating the popularity of digital content. For example, relying only on external media data may overestimate or underestimate the popularity of some digital content that is active in on-chain transactions but has low social media exposure; while focusing only on on-chain transaction data may ignore the potential driving effect of social media heat on the market value of digital content.

[0004] On the other hand, the on-chain transaction data perspective focuses on analyzing the transaction data of digital content on the blockchain network, such as transaction volume, transaction frequency, and transaction amount, to evaluate the market activity and value of digital content. However, this method also has its shortcomings, as it cannot fully capture the spread and influence of digital content on social media, and it does not include security incidents as an important factor in the analysis. In the blockchain field, security incidents such as smart contract vulnerabilities and malicious attacks occur frequently and have a significant impact on the popularity and market value of digital content, but existing technologies have not effectively addressed this challenge.

[0005] Currently, there is no unified and standard quantitative method for evaluating the popularity of on-chain digital content in the market, and market participants have strong subjectivity and uncertainty when judging content value. Traditional evaluation methods rely heavily on social media attention, sales data, etc., but in the on-chain environment, such methods often cannot fully reflect the real performance of digital content. In addition, in recent years, several incidents of digital content project blowout, contract vulnerability exploitation, and project running away have severely affected market performance and user trust, fully demonstrating the significant coupling relationship between "security" and "popularity".

[0006] On the one hand, the security vulnerabilities of smart contracts (such as re-entrant attacks, self-destruct functions, etc.) will directly weaken users' confidence in the project, resulting in a decline in transaction volume and willingness to hold; on the other hand, negative public opinion from social platforms will also exacerbate market sentiment fluctuations, triggering a rapid decline in popularity. More importantly, once a high-risk security event (such as project running, attack events, media exposure, etc.) occurs, the market heat of related digital content will usually present an "exponential collapse".

[0007] Therefore, it is particularly important to develop a chain digital content popularity quantification modeling method that not only integrates multi-dimensional indicators but also introduces a security event intervention mechanism. This method not only meets the actual needs of market participants to accurately perceive the value of digital content, but also provides a more comprehensive and dynamic monitoring means for blockchain regulatory departments, effectively identifying potential risks, suppressing malicious speculation, and improving market transparency and credibility. SUMMARY

[0008] In view of the deficiencies of the prior art, the present application precisely focuses on the gap in the current evaluation of chain digital content popularity, and proposes a multi-segment Sigmoid chain digital content popularity modeling method based on full risk adjustment. This method breaks through the limitations of traditional analysis frameworks and realizes the deep integration and comprehensive analysis of multi-dimensional data.

[0009] The present application is realized by the following technical solutions: A multi-segment Sigmoid chain digital content popularity modeling method based on security risk adjustment, which specifically includes the following steps: (1) Data acquisition stage, including the following sub-steps: (1.1) Use diversified platform API interfaces, including but not limited to blockchain browsers and decentralized trading markets, to efficiently capture chain data; set the collection period to For each , the transaction data at time t is represented as: ; The data fields of transaction data represent total transaction volume, total sales volume, average transaction price, total number of owners, market total value, minimum transaction price, time period transaction volume, time period transaction volume difference, time period transaction volume change rate, time period sales volume, time period sales volume difference, time period average transaction price, total supply, creation date; (1.2) Using Selenium technology, integrating multi-strategy anti-crawling measures, deeply crawling information from social platforms, collecting comprehensive social media interaction data related to the collection by simulating real user interaction and fine webpage analysis, ensuring the depth and breadth of data collection; for each digital content collection , social media data on time includes the following fields: ; These data fields respectively represent the number of Twitter posts, the number of Twitter followers, the number of Twitter media, the number of Discord members, the number of Discord online users, the number of Instagram media, and the number of Instagram followers; (1.3) The collected raw data comprehensively covers multiple dimensions of market performance and social media influence of on-chain digital content, including, while integrating key interaction indicators of social platforms, including the number of tweets, the number of followers, the number of media shares, and the number of online users, to evaluate community activity and the social communication power of content; also includes basic information such as creation date, symbol classification, and collection date, ensuring data structuring and time consistency, providing an information base for subsequent analysis; (1.4) Smart contract source code collection: By accessing the open API of the blockchain browser, for the unique identifier Slug of each on-chain digital content, the verified smart contract source code corresponding to its deployment address is obtained and stored as a structured code file, which is used for subsequent vulnerability static analysis; (1.5) Public opinion text data collection: Combined with Selenium + API crawling strategy, user comments, posts, and tweets related to the specified digital content project are collected from social platforms; high-relevance content is filtered through keyword matching and stored by time window for subsequent sentiment recognition; the keyword matching includes Slug, contract address, and project name; (1.6) Security event record collection: Using the interfaces or announcements provided by third-party security monitoring platforms, historical security event data is collected, including event type, occurrence time, affected project, attack method, and severity level; supplemented by publicly disclosed information sources; all events and digital content projects are associated and matched by address or keyword; the third-party security monitoring platforms include but are not limited to data aggregation and analysis platform Dune, smart contract audit agency Cerik, blockchain security research laboratory PeckShield, and on-chain threat detection system ChainArgos; (2) Data preprocessing stage, including the following sub-steps: (2.1) The collected on-chain transaction data and social media data Merging, the merging operation is carried out according to the set unique identifier slug; suppose the merged data is : ; (2.2) Due to the different units and dimensions of different data indicators, standardization processing is also needed after merging; suppose the standardized data is : ; wherein, denotes the merged original data, denotes the mean value of the data, denotes the standard deviation of the data; (2.3) Using principal component analysis PCA to reduce the dimension of the standardized data, and extracting three principal components; (3) Identification and screening of key turning points, the specific process is as follows: (3.1) Using moving average difference method to efficiently identify turning points; (3.2) Based on the identified key turning points, set some rules to screen them; (4) Multi-segment Sigmoid function quantitative modeling stage, using multi-segment Sigmoid function quantitative modeling method to quantify the on-chain digital content; (5) Security event risk analysis stage, in order to systematize the security risk factors into the modeling process, by setting three parallel sub-modules, respectively processing contract security, public opinion risk and security events that have occurred, and through a unified scoring mechanism to adjust the security of the modeling results: (5.1) Smart contract security score; (5.2) Public opinion sentiment analysis; (5.3) Security event response; (6) Popularity comprehensive scoring stage; Finally, for each collection, combining the original multi-segment Sigmoid function modeling score , the final security risk perception popularity score of on-chain digital content is calculated; Through the above steps, the quantitative modeling process of on-chain digital content popularity is completed, and the popularity of on-chain digital content is comprehensively, accurately and dynamically evaluated, and the change trend of popularity in different time periods is obtained.

[0010] In particular, the raw data collected in step (1.3) comprehensively covers multiple dimensions of the market performance and social media influence of the on-chain digital content, including: unique identifier Slug for distinguishing different collection series; total transaction volume and total sales volume reflecting the scale and frequency of transaction activities; average transaction price showing the average value cognition of the market for the content; total owner quantity reflecting the breadth of the user base; market total value summarizing the overall market value; minimum transaction price marking the market bottom price; and time interval related indicators, i.e., interval transaction volume changes, for revealing short-term dynamic trends.

[0011] Further, step (2.3) uses principal component analysis (PCA) to reduce the dimensionality of the standardized data and extract three principal components; specifically including the following steps: (2.3.1) Let the standardized data matrix be with dimensions , where is the number of samples, is the number of features; and the covariance matrix is calculated as: ; (2.3.2) Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors : ; (2.3.3) Select the eigenvectors corresponding to the three largest eigenvalues to form the principal component matrix ; (2.3.4) Project the standardized data onto the principal component matrix to obtain the reduced dimensionality data : ; According to the composition and weights of these principal components, they are defined as transaction activity, value recognition, and community influence, respectively.

[0012] Further, step (3.1) uses the moving average difference method to efficiently identify turning points, specifically including the following steps: (3.1.1) Set a window size and calculate the moving average of the time series : ​ ; (3.1.2) Calculate the time series Its moving average Difference between : ; (3.1.3) The initial turning point is at The moment when significant changes occur.

[0013] Furthermore, in step (3.2), the identified key turning points are filtered by setting certain rules, specifically including the following steps: (3.2.1) Calculate the difference absolute value and standard deviation Then the confidence threshold Set as the mean plus a certain number of standard deviations: ; in, The selected inflection points satisfy the multiple parameter. ; (3.2.2) Use percentiles to set confidence thresholds For example, setting the confidence threshold to the first degree of the absolute difference. Percentiles: ; in, For the first The percentiles, the selected inflection points satisfy the following conditions .

[0014] Furthermore, step (4) employs a multi-segment Sigmoid function quantization modeling method to quantize the on-chain digital content, specifically including the following sub-steps: (4.1) First, based on the inflection points identified and selected in the previous steps, the time series is divided into several intervals; assuming the inflection point is... Divide the time series into Each interval: ; (4.2) Within each interval, quantization modeling is performed using the Sigmoid function, let the first interval be... The sigmoid function for the segment is: ; in, The indexes representing the principal components include trading activity, value recognition, and community influence; is the maximum value of the Sigmoid function, corresponding to the peak value of the first principal component in the interval; is the steepness of the curve, which determines the rising or falling speed of the curve; is the first turning point of the first principal component; is the first turning point of the first principal component; (4.3) Use the trust region reflective algorithm TRR to optimize the parameters of each segment of the Sigmoid function ; TRR algorithm is a nonlinear optimization method suitable for parameter optimization. When the upper and lower bounds of the parameters are known, the optimization goal is to minimize the error between the actual data and the fitted Sigmoid function. The mean square error MSE is usually used as the evaluation standard, and the expression is as follows: ; where, is the actual observation value, is the corresponding Sigmoid function prediction value; (4.4) Sum the results of each segment of the Sigmoid function to obtain the quantitative modeling result of a certain : ; where, is the weight of the first principal component, reflecting the relative importance of each principal component in the overall popularity.

[0015] Further, the step (5.1) smart contract security score, specifically comprising the following steps: (5.1.1) Input the smart contract source code collected from (1.4); (5.1.2) For each collection of digital content smart contract source code, use Slither static code analysis tool to run the full rule library, automatically identify reentrant attacks, integer overflow, lack of permission control, self-destruction logic and other security vulnerabilities; (5.1.3) Set the risk weight of each type of vulnerability , and each type of vulnerability is counted once, that is, plus one point. The original score of the contract risk is: ; (5.1.4) If , there are multiple serious vulnerabilities, defined as high risk; if , there are some controllable vulnerabilities, defined as medium risk; otherwise, there are only slight or no risk problems, defined as low risk. The results will be normalized in the future, that is: ; wherein is the maximum risk value in all current samples.

[0016] Further, the step (5.2) public opinion sentiment analysis specifically includes the following steps: (5.2.1) input the text content collected from step (1.5); (5.2.2) for each public opinion event corresponding to a collection, use the multi-round fine-tuned BERT sentiment classification model to classify each text as positive, neutral or negative; (5.2.3) in time window t, count the proportion of negative emotions , calculate the public opinion risk coefficient; ; wherein, is an adjustment factor set by experience, default is 0.7; support weighted by user type or community influence.

[0017] Further, the step (5.3) security event response; specifically includes the following steps: (5.3.1) the historical security events from step (1.6) collection (5.3.2) for each on-chain digital content item, if a public security event occurs within time window t, check its severity level (such as high-risk attack, system failure, community bug), and define the corresponding penalty factor: ; wherein: , respectively, represent low, medium and high risk events, represents the risk decay control parameter, the recommended value is 0.4~0.8; if there is no event, , that is, no penalty; (5.4) security factor fusion and final correction: The above three risk scores are normalized to obtain the comprehensive security risk score : ; wherein, , the sum of the weights of smart contract risk, public opinion risk and event attack; the default ratio is 0.4:0.3:0.3, and can be adjusted according to actual application.

[0018] Further, the step (6) of the on-chain digital content security risk perception popularity score; specifically obtained by the following formula: ; wherein, To synthesize the safety risk score; this formula completes the dynamic adjustment of the popularity results of risk events, and makes the model have real-time response under the driving of safety events.

[0019] The beneficial effects of the present application are as follows: The present application is targeted at innovation, realizes multi-dimensional data deep fusion, organically combines external media data, on-chain transaction data and safety event data, and fully presents the real face of the digital content market, avoiding evaluation deviation of a single data dimension. For the first time, safety events are included in the evaluation system, with the help of static code analysis, BERT model sentiment analysis and safety event classification, the safety score is accurately calculated, and the potential impact of safety events on popularity is fully considered to improve the evaluation accuracy. The innovative introduction of sentiment recognition based on deep learning model (BERT) breaks through the limitations of traditional public opinion analysis and accurately captures public attitudes and trust tendencies. A dynamic risk adjustment mechanism is constructed to immediately suppress and correct the popularity score when safety events occur, ensuring that the evaluation keeps up with market fluctuations and meets the high timeliness requirements. The modeling structure is optimized, and the use of multi-segment Sigmoid function model not only accurately captures the stage fluctuations and key turning points of popularity, but also models the entire life cycle of digital content, enhancing the adaptability and flexibility of the model, and providing a more comprehensive, accurate, dynamic and practical popularity evaluation tool for market analysis, investment decision-making and supervision of on-chain digital content. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The present application is targeted at innovation, realizes multi-dimensional data deep fusion, organically combines external media data, on-chain transaction data and safety event data, and fully presents the real face of the digital content market, avoiding evaluation deviation of a single data dimension. For the first time, safety events are included in the evaluation system, with the help of static code analysis, BERT model sentiment analysis and safety event classification, the safety score is accurately calculated, and the potential impact of safety events on popularity is fully considered to improve the evaluation accuracy. The innovative introduction of sentiment recognition based on deep learning model (BERT) breaks through the limitations of traditional public opinion analysis and accurately captures public attitudes and trust tendencies. A dynamic risk adjustment mechanism is constructed to immediately suppress and correct the popularity score when safety events occur, ensuring that the evaluation keeps up with market fluctuations and meets the high timeliness requirements. The modeling structure is optimized, and the use of multi-segment Sigmoid function model not only accurately captures the stage fluctuations and key turning points of popularity, but also models the entire life cycle of digital content, enhancing the adaptability and flexibility of the model, and providing a more comprehensive, accurate, dynamic and practical popularity evaluation tool for market analysis, investment decision-making and supervision of on-chain digital content. DETAILED DESCRIPTION

[0021] The present application is targeted at innovation, realizes multi-dimensional data deep fusion, organically combines external media data, on-chain transaction data and safety event data, and fully presents the real face of the digital content market, avoiding evaluation deviation of a single data dimension. For the first time, safety events are included in the evaluation system, with the help of static code analysis, BERT model sentiment analysis and safety event classification, the safety score is accurately calculated, and the potential impact of safety events on popularity is fully considered to improve the evaluation accuracy. The innovative introduction of sentiment recognition based on deep learning model (BERT) breaks through the limitations of traditional public opinion analysis and accurately captures public attitudes and trust tendencies. A dynamic risk adjustment mechanism is constructed to immediately suppress and correct the popularity score when safety events occur, ensuring that the evaluation keeps up with market fluctuations and meets the high timeliness requirements. The modeling structure is optimized, and the use of multi-segment Sigmoid function model not only accurately captures the stage fluctuations and key turning points of popularity, but also models the entire life cycle of digital content, enhancing the adaptability and flexibility of the model, and providing a more comprehensive, accurate, dynamic and practical popularity evaluation tool for market analysis, investment decision-making and supervision of on-chain digital content.

[0022] The present application is targeted at innovation, realizes multi-dimensional data deep fusion, organically combines external media data, on-chain transaction data and safety event data, and fully presents the real face of the digital content market, avoiding evaluation deviation of a single data dimension. For the first time, safety events are included in the evaluation system, with the help of static code analysis, BERT model sentiment analysis and safety event classification, the safety score is accurately calculated, and the potential impact of safety events on popularity is fully considered to improve the evaluation accuracy. The innovative introduction of sentiment recognition based on deep learning model (BERT) breaks through the limitations of traditional public opinion analysis and accurately captures public attitudes and trust tendencies. A dynamic risk adjustment mechanism is constructed to immediately suppress and correct the popularity score when safety events occur, ensuring that the evaluation keeps up with market fluctuations and meets the high timeliness requirements. The modeling structure is optimized, and the use of multi-segment Sigmoid function model not only accurately captures the stage fluctuations and key turning points of popularity, but also models the entire life cycle of digital content, enhancing the adaptability and flexibility of the model, and providing a more comprehensive, accurate, dynamic and practical popularity evaluation tool for market analysis, investment decision-making and supervision of on-chain digital content.

[0023] For modeling the evolution trend of popularity, the technical personnel can use numpy to identify key inflection points by means of moving average difference method (MAD), and use scipy.optimize.curve_fit to build a multi-segment Sigmoid fitting function to model each stage of the principal component. In order to improve the fitting accuracy and parameter stability, the trust region reflection algorithm (TRR) is introduced to nonlinearly optimize the Sigmoid parameters, and finally the structured model output describing the popularity trend of on-chain digital content is obtained.

[0024] In terms of security event risk analysis, the technical personnel can use the API of a blockchain browser (such as Etherscan) to obtain the source code of the smart contract corresponding to each on-chain digital content, and call the Slither tool to perform static analysis on the contract source code, automatically identify common vulnerabilities and score and classify them, and generate a contract security risk score. At the same time, through the interface of Selenium and social media, text comment data related to digital content is collected, and a deep learning framework (such as transformers) is used to load a BERT model for sentiment classification, and the proportion of negative sentiment is counted to reflect the user trust trend.

[0025] In addition, the technical personnel can use the API or announcement data source provided by third-party security event platforms (such as data aggregation analysis platform Dune, smart contract audit agency Cerik, blockchain security research laboratory PeckShield, and on-chain threat detection system ChainArgos) to collect historical security event information related to digital content, and set an index penalty function according to the severity (high, medium, and low) to build a risk suppression factor. Finally, the contract risk score, public opinion sentiment coefficient, and event penalty function are weighted and fused to generate a comprehensive security risk factor, which is used to dynamically modify the results of the Sigmoid model.

[0026] The entire process consists of data collection, feature extraction, trend modeling, and risk regulation, forming a complete closed loop, with high realizability and flexible expansion capability, suitable for various on-chain digital content popularity quantification evaluation scenarios.

[0027] As shown in Figure 1 A multi-dimensional index and multi-segment Sidmoid function-based on-chain digital content quantification modeling method, the specific implementation includes five stages: Stage One: Data Collection Stage Stage Two: Data Preprocessing Stage Stage Three: Tippoing Point Identification and Screening Stage Stage Four: Multi-segment Sigmoid Function Quantification Modeling Stage Stage Five: Security Event Risk Analysis Stage Phase six, popularity comprehensive score The phase one includes the following steps: (1.1) Efficiently scraping on-chain data using diversified platform API interfaces, including but not limited to blockchain browsers and decentralized trading markets. Set the collection period to (once a day), for each , the transaction data at time t can be represented as: ; These data fields respectively represent total transaction volume, total sales volume, average transaction price, total number of owners, market total value, lowest transaction price, period transaction volume, period transaction volume difference, period transaction volume change rate, period sales volume, period sales volume difference, period average transaction price, total supply, creation date.

[0028] (1.2) Use advanced crawler technologies such as Selenium, integrate multi-strategy anti-crawling measures, deeply crawl information from social platforms such as Twitter, Instagram, Discord, etc. Through simulating real user interaction and fine webpage analysis, comprehensively collect social media interaction data related to collection (a set of digital content), and ensure the depth and breadth of data collection. For each , the social media data at time includes the following fields: ; These data fields respectively represent the number of Twitter posts, the number of Twitter followers, the number of Twitter media, the number of Discord members, the number of Discord online users, the number of Instagram media, and the number of Instagram followers.

[0029] (1.3) The collected raw data comprehensively covers multiple dimensions of market performance and social media influence of on-chain digital content (especially collections), including: unique identifier Slug used to distinguish different collection series; total transaction volume ( ) and total sales volume ( ) reflect the scale and frequency of transaction activities; average transaction price ( ) shows the average value cognition of the market for this content; total number of owners ( ) reflects the breadth of user base; market total value ( ) summarizes the overall market value; lowest transaction price ( ​) Marking the bottom price of the market. Time interval related indicators, such as interval trading volume changes, reveal short-term dynamic trends. At the same time, the data also integrates key interaction indicators from social platforms such as Twitter, Instagram, Discord, etc., such as the number of tweets, the number of fans, the number of media shares, and the number of online people, to evaluate the community activity and the social communication power of the content. In addition, it also includes basic information such as creation date, symbol classification, and collection date, ensuring the structure and time consistency of the data, providing a rich and detailed information base for in-depth analysis.

[0030] The phase two includes the following steps: (2.1) Merge the collected on-chain transaction data and social media data according to the collection unique identifier slug. Let the merged data be :

[0031] (2.2) Since the units and dimensions of different data indicators are different, standardization processing is needed after merging. Let the standardized data be :

[0032] where represents the original data after merging, represents the mean of the data, represents the standard deviation of the data.

[0033] (2.3) Use principal component analysis (PCA) to reduce the dimension of the standardized data and extract three principal components, which includes the following steps: (2.3.1) Let the standardized data matrix be , with dimensions , where is the number of samples, is the number of features. The covariance matrix is calculated as: ; (2.3.2) Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and eigenvectors : ; (2.3.3) Select the eigenvectors corresponding to the largest three eigenvalues to form the principal component matrix .

[0034] (2.3.4) Standardizing the data Projecting to principal component matrix , we get the reduced dimension data : ; According to the composition of these principal components and their weights, they are defined as Transaction Activity, Value Recognition and Community Influence, respectively The phase three includes the following steps: (3.1) Using moving average difference method to efficiently identify the tipping point, including the following steps: (3.1.1) Set a window size , calculate the moving average of time series : : ; (3.1.2) Calculate the difference between time series and its moving average : : ; (3.1.3) The preliminary tipping point is the moment when changes significantly.

[0035] (3.2) Based on the identified tipping point (key turning point), filter it by setting certain rules: (3.2.1) Calculate the absolute value and standard deviation of the difference , then set the confidence threshold to the mean plus several times the standard deviation: ; Where, is the multiple parameter, and the filtered tipping point satisfies .

[0036] (3.2.2) Use percentile to set the confidence threshold , for example, set the confidence threshold to the th percentile of the absolute value of the difference: ; Where, is the th percentile, and the filtered tipping point satisfies .

[0037] The step four includes the following steps: (4.1) First, the time series is divided into several intervals according to the identified and filtered turning points in the previous steps. Suppose the turning points are , the time series is divided into intervals: ; (4.2) In each interval, a Sigmoid function is used for quantification modeling. Let the Sigmoid function of the th interval be: ; where represents the index of the principal component (e.g., transaction activity, value recognition, and community influence), is the maximum value of the Sigmoid function, corresponding to the peak value of the th principal component in the interval; is the steepness of the curve, determining the rising or falling speed of the curve; is the th turning point of the th principal component.

[0038] (4.3) The Trust Region Reflective (TRR) algorithm is used to optimize the parameters of each Sigmoid function. TRR is a nonlinear optimization method suitable for parameter optimization, especially when the upper and lower bounds of the parameters are known. The optimization goal is to minimize the error between the actual data and the fitted Sigmoid function, usually using Mean Squared Error (MSE) as the evaluation standard: ; where is the actual observation value, is the corresponding Sigmoid function prediction value.

[0039] (4.4) The results of each Sigmoid function are summed up to obtain the quantification modeling result of a certain : ; where is the weight of the th principal component, reflecting the relative importance of each principal component in the overall popularity.

[0040] (5) Security event risk analysis stage, in order to systematize the security risk factors into the modeling process, the present application sets up three parallel sub-modules, which respectively deal with contract security, public opinion risk and security events that have occurred, and adjust the security of the modeling results through a unified scoring mechanism: (5.1) Smart contract security scoring module, specifically comprising the following steps: (5.1.1) The input comes from the smart contract source code collected in (1.4); (5.1.2) For each type of collection corresponding to the digital content smart contract source code, use Slither static code analysis tool to run the full rule base, automatically identify security vulnerabilities such as reentrant attack, integer overflow, lack of permission control, self-destruction logic, etc. (5.1.3) Set the risk weight of each type of vulnerability , each type of vulnerability is counted once, and the original score of the contract risk is:

[0041] (5.1.4) If , there are multiple serious vulnerabilities, defined as high risk; if , there are some controllable vulnerabilities, defined as medium risk; otherwise, there are only slight or no risk problems, defined as low risk. The results will be normalized in the following steps: ; Among them is the maximum risk value in the current all samples.

[0042] (5.2) Public opinion sentiment analysis module, specifically comprising the following steps: (5.2.1) The input comes from the text content collected in step (1.5); (5.2.2) For each type of collection corresponding to the public opinion event, use the multi-round fine-tuning BERT sentiment classification model to classify each text as positive, neutral or negative; (5.2.3) Take the time window t as the unit, calculate the proportion of negative emotions , and calculate the public opinion risk coefficient; ; Among them, is an experience setting adjustment factor (default is 0.7). Support weighted by user type or community influence.

[0043] (5.3) Security event response module, specifically comprising the following steps: (5.3.1) The input comes from the historical security events collected in step (1.6) (5.3.2) For each chain digital content item, if a public security event occurs within the time window t, check its severity level (such as high-risk attack, system failure, community bug), and define the corresponding penalty factor: ; Where: , respectively, represent low, medium and high risk events, represents the risk attenuation control parameter, the recommended value is 0.4~0.8; if there is no event, (no penalty).

[0044] (5.4) Security factor fusion and final correction: The above three kinds of risk scores are normalized and synthesized to obtain the comprehensive security risk score : ; Where, , respectively, are the weights of smart contract risk, public opinion risk, and event attack. The default recommended ratio is 0.4:0.3:0.3, which can be adjusted according to actual application.

[0045] (6) Popularity comprehensive score Finally, for each collection, the original multi-section Sigmoid function is modeled to score , which is obtained by the following formula: The security risk perception popularity score of the on-chain digital content is: ;

[0046] The formula realizes the dynamic adjustment of the popularity result by the risk event, so that the model has the real-time response ability under the driving of the security event, and improves the description ability and modeling accuracy of the real market trend.

[0047] In summary, the present application introduces the moving average difference method (MAD) to identify key turning points, and sets thresholds based on mean and standard deviation or percentile, effectively filtering out real market change signals. The data collection frequency is updated periodically every day, ensuring the continuity and real-time nature of the data. In terms of advantages and effects, this method can timely capture market change signals, significantly improving the timeliness and sensitivity of the model, overcoming the lag problem of existing technologies in capturing dynamic changes, so that the model can more accurately reflect the real-time changes of the market. In terms of the accuracy of quantitative modeling, the present application uses a segmented Sigmoid function for quantitative modeling, fitting each filtered interval separately, and using the trust region reflection algorithm (TRR) to optimize the parameters of each Sigmoid function. Finally, by weighted sum of each Sigmoid function, the popularity quantitative result of each collection is obtained. This segmented modeling method improves the accuracy of the model, which can more accurately capture the changes of the popular trend, and overcomes the shortcomings of the rough quantitative modeling in the prior art, so that the model result has more practical value and application prospect. In addition, the present application first introduces a systematic security event risk analysis module, including contract vulnerability static analysis (based on Slither), social public opinion sentiment recognition (based on BERT model) and historical security event level penalty mechanism (exponential penalty function). The three types of information are fused to calculate the security factor, and the original modeling result is dynamically adjusted. This mechanism builds a "credibility-emotion perception-risk impact" closed-loop feedback system, which significantly enhances the model's perception and quantitative ability of technical security, user trust trend and risk propagation effect. Compared with the traditional method, the present application can actively suppress the popularity score to accurately reflect the real market signal, filling the gap in the "risk adjustment dimension" of the prior art.

[0048] The present application integrates multi-source data collection, standardized feature extraction, dynamic trend fitting, security event perception and score adjustment, and other key technical paths, and builds a full-process, closed-loop modeling system from data input to score output. It provides a "comprehensive, multi-dimensional, security-aware, and dynamically adaptive" popularity modeling tool, which has higher accuracy, stronger dynamic adaptability and deeper risk reflection ability than existing methods, providing a high-credibility decision support tool for content platforms, investment institutions and regulatory authorities, and has wide application value and industry leading nature. Therefore, the present application has significant innovations and optimizations in data sources, processing methods, dynamic change capture and quantitative modeling, and can effectively overcome the shortcomings of the prior art, providing a comprehensive, accurate and dynamic popularity quantitative tool.

[0049] The above merely describes preferred embodiments of the present application, but is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0050] The above embodiments are only used to illustrate the design ideas and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the present application and implement it, and the protection scope of the present application is not limited to the above embodiments. Therefore, any equivalent change or modification made according to the disclosed principles and design ideas of the present application is within the protection scope of the present application.

Claims

1. A method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment, characterized in that, The method specifically includes the following steps: (1) Data acquisition stage, including the following sub-steps: (1.1) Utilize diverse platform API interfaces, including but not limited to blockchain explorers and decentralized exchanges, to efficiently capture on-chain data; set the collection period as follows: For each Transaction data at time t It can be represented as: ； These data fields represent total transaction volume, total sales volume, average transaction price, total number of owners, total market value, minimum transaction price, transaction volume during a period, transaction volume difference during a period, transaction volume change rate during a period, sales volume during a period, sales volume difference during a period, average transaction price during a period, total supply, and creation date, respectively. (1.2) Employing Selenium technology and integrating multi-strategy anti-crawling measures, information is deeply crawled from social media platforms. By simulating real user interactions and performing detailed webpage analysis, comprehensive social media interaction data related to the collection is collected to ensure the depth and breadth of data collection; for each digital content collection In time Social media data Includes the following fields: ； These data fields represent the number of Twitter posts, the number of Twitter followers, the number of Twitter posts, the number of Discord members, the number of Discord users online, the number of Instagram posts, and the number of Instagram followers, respectively. (1.3) The collected raw data comprehensively covers multiple dimensions of on-chain digital content (market performance and social media influence), Meanwhile, the data also integrates key interaction metrics from social media platforms, such as the number of tweets, number of followers, media shares, and number of online users, to assess community activity and the social reach of content; it also includes basic information such as creation date, symbol classification, and collection date to ensure the structure and temporal consistency of the data, providing an information foundation for subsequent analysis; (1.4) Smart contract source code collection: By accessing the blockchain explorer's open API, for each unique identifier Slug of on-chain digital content, obtain the verified smart contract source code corresponding to its deployment address, store it as a structured code file, and use it for subsequent static vulnerability analysis; (1.5) Public opinion text data collection: Combining the Selenium + API crawling strategy, collect user comments, posts and tweets related to the specified digital content project from social platforms; filter highly relevant content by keyword matching, classify and store it according to time windows, and use it for sentiment recognition later; the keyword matching includes Slug, contract address and project name; (1.6) Security Incident Record Collection: Collect historical security incident data using interfaces or announcements provided by third-party security monitoring platforms, including fields such as incident type, occurrence time, affected projects, attack methods, and severity level; supplemented by publicly disclosed information sources; all incidents and digital content projects are matched and associated through addresses or keywords; the third-party security monitoring platforms include, but are not limited to, the data aggregation and analysis platform Dune, the smart contract auditing agency Cerik, the blockchain security research laboratory PeckShield, and the on-chain threat detection system ChainArgos; (2) Data preprocessing stage, including the following sub-steps: (2.1) Collect on-chain transaction data and social media data The merging process is performed based on the set's unique identifier, slug; let the data to be merged be... : ； (2.2) Since the units and dimensions of different data indicators are different, standardization is still required after merging; let the standardized data be... : ； in, This represents the original data after merging. This represents the mean of the data. The standard deviation of the data; (2.3) Principal component analysis (PCA) was used to reduce the dimensionality of the standardized data and extract three principal components; (3) The identification and screening stage of key turning points, specifically the following process: (3.1) The moving average difference method is used to efficiently identify inflection points; (3.2) Based on the identified key turning points, a certain rule is set to filter them; (4) In the multi-segment Sigmoid function quantization modeling stage, the on-chain digital content is quantized using the multi-segment Sigmoid function quantization modeling method; (5) In the security incident risk analysis stage, in order to systematically incorporate security risk factors into the modeling process, three parallel sub-modules are set up to handle contract security, public opinion risk and security incidents that have occurred, respectively, and a unified scoring mechanism is used to adjust the security of the modeling results: (5.1) Smart contract security score; (5.2) Public opinion sentiment analysis; (5.3) Security incident response; (6) Popularity comprehensive scoring stage Finally, for each collection, a score is modeled by combining the original multi-segment Sigmoid function. Finally, the popularity score of the perceived security risk of on-chain digital content is calculated. Through the above steps, the quantitative modeling process of the popularity of on-chain digital content has been completed. This model can comprehensively, accurately and dynamically evaluate the popularity of on-chain digital content and reflect the changing trend of popularity over different time periods.

2. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, The raw data collected in step (1.3) comprehensively covers multiple dimensions of on-chain digital content (market performance and social media influence), specifically including: a unique identifier Slug used to distinguish different collection series; total transaction volume ( ) and total sales volume ( ) reflects the scale and frequency of trading activities; average transaction price ( This indicates the market's average perceived value of the content; the total number of owners ( This reflects the breadth of the user base; total market value ( Summarize the overall market value; minimum transaction price ( This marks the bottom price level of the market. Time-related indicators, such as changes in trading volume over time, reveal short-term dynamic trends.

3. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, Step (2.3) uses Principal Component Analysis (PCA) to reduce the dimensionality of the standardized data and extract three principal components; specifically, it includes the following steps: (2.3.1) Let the standardized data matrix be... Its dimensions are ,in For the sample size, The number of features. Covariance matrix. The calculation formula is: ； (2.3.2) For the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues. and eigenvectors : ； (2.3.3) Select the eigenvectors corresponding to the three largest eigenvalues ​​to form the principal component matrix. ; (2.3.4) Standardize the data Projected onto principal component matrix The data obtained after dimensionality reduction is obtained. : ； Based on the composition and weights of these principal components, they are defined as trading activity, value recognition, and community influence, respectively.

4. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, Step (3.1) uses the moving average difference method to efficiently identify inflection points, specifically including the following steps: (3.1.1) Set a window size Calculate time series moving average : ； (3.1.2) Calculate the time series Its moving average Difference between : ； (3.1.3) The initial turning point is at The moment when significant changes occur.

5. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, In step (3.2), the identified key turning points are filtered using certain rules, specifically including the following steps: (3.2.1) Calculate the difference absolute value and standard deviation Then the confidence threshold Set as the mean plus a certain number of standard deviations: ； in, The selected inflection points satisfy the multiple parameter. ; (3.2.2) Use percentiles to set confidence thresholds For example, setting the confidence threshold to the first degree of the absolute difference. Percentiles: ； in, For the first The percentiles, the selected inflection points satisfy the following conditions .

6. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, The method of multi-segment Sigmoid function quantization modeling in step (4) quantizes the digital content on the chain, specifically including the following sub-steps: (4.1) First, based on the inflection points identified and filtered in the previous steps, the time series is divided into several intervals. Assume the inflection point is... Divide the time series into Each interval: ; (4.2) Within each interval, quantization modeling is performed using the Sigmoid function, let the first interval be... The sigmoid function for the segment is: ； in, The indexes representing the principal components include trading activity, value recognition, and community influence; It is the maximum value of the Sigmoid function, corresponding to the th in this interval. The peak value of each principal component; The steepness of the curve determines the speed at which the curve rises or falls. It is the first The first principal component A turning point; (4.3) Use the Trust Region Reflection (TRR) algorithm to optimize the parameters of each segment of the Sigmoid function. The TRR algorithm is a nonlinear optimization method suitable for parameter optimization, especially when the upper and lower bounds of the parameters are known. The optimization objective is to minimize the error between the actual data and the fitted sigmoid function, typically using mean squared error (MSE) as the evaluation metric. ； in, These are actual observed values. The predicted value for the corresponding Sigmoid function; (4.4) Sum the results of each segment of the Sigmoid function to obtain a certain... Quantitative modeling results : ； in, For the first The weights of each principal component reflect the relative importance of each principal component in the overall popularity.

7. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, Step (5.1), smart contract security scoring, specifically includes the following steps: (5.1.1) Input is the smart contract source code collected from (1.4); (5.1.2) For the source code of the digital content smart contract corresponding to each collection, use the Slither static code analysis tool to run the full rule library and automatically identify security vulnerabilities such as reentrancy attacks, integer overflows, lack of access control, and self-destruct logic. (5.1.3) Set risk weights for vulnerability types One point is added for each type of vulnerability recorded. The original contract risk score is: ； (5.1.4) If If so, there are multiple serious vulnerabilities, and it is defined as high risk; if If the value is 0, then there are some controllable vulnerabilities, defined as medium risk; otherwise, there are only minor or no risk issues, defined as low risk. The results will then be normalized: ； in This represents the maximum risk value among all current samples.

8. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, Step (5.2) of public opinion sentiment analysis specifically includes the following steps: (5.2.1) Input the text content collected in step (1.5); (5.2.2) For each collection corresponding to a public opinion event, a multi-round fine-tuned BERT sentiment classification model is used to classify each text as positive, neutral or negative; (5.2.3) Calculate the percentage of negative emotions using time window t as the unit. Calculate the public opinion risk coefficient; ； in, Adjustment factor set for experience (default is 0.7). Supports weighting by user type or community influence.

9. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, Step (5.3) Security Incident Response; specifically includes the following steps: (5.3.1) Historical security events collected in step (1.6) (5.3.2) For each on-chain digital content project, if a public security incident occurs within the time window t, its severity level (e.g., high-risk attack, system failure, community bug) shall be checked, and a corresponding penalty factor shall be defined: ； in: , respectively representing low, medium, and high risk events. This represents the risk decay control parameter; a recommended value is 0.4 to 0.

8. If no events occur, then... That is, no punishment; (5.4) Safety factor fusion and final correction: The three risk scores mentioned above are normalized and combined to obtain a comprehensive safety risk score. : ； in, This is the sum of the weights of smart contract risks, public opinion risks, and incident attacks; the default ratio is 0.4:0.3:0.3, which can be adjusted according to actual applications.

10. The method for modeling the popularity of digital content on a multi-segment Sigmoid chain based on security risk adjustment according to claim 1, characterized in that, The security risk perception popularity score of on-chain digital content in step (6) is obtained through the following formula: ； in, This formula provides a comprehensive security risk score; it dynamically adjusts the prevalence results based on risk events and enables the model to respond in real time to security events.

Images