Identification early warning and law prompting method based on big data

Abstract

The invention provides an identification early warning and law prompting method based on big data, and the method comprises the steps: collecting the multi-dimensional behavior data of a user in real time, including identity features, historical transaction prices and real-time environment variables; performing space-time association processing on the data to generate a space-time aligned feature tensor; on the basis of user price sensitivity grading, differential weighted fusion is carried out on the feature tensors, and a classification feature matrix is generated; a dynamic deviation index is calculated, a price rationality comprehensive score is generated, and early warning is triggered by a low score; inputting a legal clause matching engine to generate an early warning report; and executing a multi-level intervention strategy according to the early warning level. The method can help a user identify potential price differences, provides technical support for a related platform, and enables the user to find and process price abnormity behaviors more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis and artificial intelligence. More specifically, the present invention relates to a method for identification warning and legal prompt based on big data. Background Art

[0002] With the booming development of e-commerce, online shopping has become an indispensable part of people's daily lives. Merchants use big data analysis to optimize pricing strategies, enhance user experience, and improve operational efficiency. However, this data-driven pricing strategy has also raised some problems. Among them, there are merchants who use the consumer personal information they possess to implement higher prices or other unfavorable trading conditions for old users or users with high consumption capabilities in order to obtain more profits. Such behavior not only damages the legitimate rights and interests of consumers.

[0003] The existing pricing strategies and regulatory mechanisms mainly rely on the self-discipline of merchants and the ex post complaints of consumers. Merchants usually adjust prices based on data such as users' purchase history and browsing behavior, but such adjustments often lack transparency and are difficult for consumers to detect. When dealing with such feedback, due to the lack of real-time monitoring and effective technical means, it is often difficult to discover and solve problems in a timely manner. In addition, the existing technical means mainly focus on user behavior analysis and price monitoring, but these methods can often only detect price fluctuations and cannot accurately judge whether there is price misrepresentation behavior, let alone provide strong evidence and warning prompts. Summary of the Invention

[0004] The present invention provides a method for identification warning and legal prompt based on big data, including: S1. Real-time collect multi-dimensional user behavior data, including user identity characteristic data, historical transaction price sequences, and real-time environmental variable data, where the user identity characteristic data at least includes device fingerprint characteristics, membership level labels, and consumption ability evaluation values, and the consumption ability evaluation value is obtained by weighted calculation of the user's historical order amount and consumption frequency; S2. Perform spatio-temporal correlation processing on the multi-dimensional behavior data, and generate a spatio-temporally aligned feature tensor through abnormal transaction time point detection and cross-platform user identity mapping, where the cross-platform user identity mapping is based on double verification of device fingerprint similarity and consumption behavior consistency; S3. Based on the user price sensitivity classification results, perform differential weighted fusion on the feature tensor to generate a user group classification feature matrix, where the historical transaction price fluctuation threshold of the high price sensitivity user group is reduced by 20%-30%; S4. Calculate the dynamic deviation index of the real-time transaction price relative to the user's historical benchmark price, and fuse the real-time supply and demand index and the promotion intensity parameter to generate a comprehensive price rationality score, and trigger an early warning mechanism when the score is lower than the preset risk level; S5. Input the early warning scenario feature vector into the legal clause matching engine to generate an early warning report containing the legal basis and penalty clauses for anti-price discrimination. The legal clause matching engine is implemented through legal text semantic parsing and scenario feature vector similarity calculation; S6. Execute a multi-level intervention strategy according to the early warning report level, including pushing real-time price comparison information to the user terminal, sending an interface throttling instruction to the enterprise risk control system, and submitting a risk record data packet to the supervision platform.

[0005] Further, step S2 includes: S21. Detect abnormal trading time points using a dynamic time window. When it is detected that the fluctuation range of the trading price within the time window exceeds 3 times the standard deviation of the same period in history, compensate and correct the original timestamp. The correction formula is: ;

[0006] where, represents the trading price at time t, represents the average price within the window period, is the dynamic compensation coefficient and , is the sign function; S22. Construct a cross-platform user identity mapping graph. By calculating the similarity of device fingerprint hash values and the matching degree of consumption behavior sequences, generate a user identity confidence score. When the confidence score is greater than 0.8, it is determined as the same user entity; S23. Reconstruct the time-space aligned data into a three-dimensional feature tensor. The dimensions include the number of users, the number of time slices, and the number of feature channels. The feature channels at least include the price volatility, the change amount of consumption frequency, and the environmental variable impact factor.

[0007] Further, the differential weighted fusion in step S3 includes: S31. Calculate the user price sensitivity index, and comprehensively evaluate it by analyzing three dimensions of the price comparison response time, coupon usage efficiency, and alternative product selection tendency in the user's historical orders. The weight ratio of the alternative product selection tendency is not less than 40%; S32. Construct a user group classification model. Divide the users with the top 10% of the price sensitivity index into the high-sensitivity group, the middle 30% into the medium-sensitivity group, and the rest into the low-sensitivity group. Set different price fluctuation tolerance coefficients for different groups; S33. Perform group weighted pooling processing on the feature tensor. The weights of the feature channels of the high-sensitivity group are increased by 1.5 - 2 times, and the weights of the low-sensitivity group are reduced to 0.7 - 0.9 times to generate a classification feature matrix with group discrimination.

[0008] Further, step S4 includes: S41. Generate the user's historical benchmark price curve, and fuse the user's past K transaction prices using the exponential decay weighting method, where the weight ratio of the transaction prices in the most recent 30 days is not less than 60%; S42. Calculate the dynamic price deviation index. By comparing the absolute difference between the real-time transaction price and the benchmark price, and superimposing the influencing factor of the deviation of the real-time supply and demand index, generate a normalized deviation score. The specific formula is: ;

[0009] where, is the benchmark price, represents the deviation degree of the real-time supply and demand index from the historical benchmark value, is the benchmark value of the supply and demand index; S43. Construct a price rationality scoring model. When the deviation index exceeds the tolerance threshold of the user's group, start the price rationality degradation mechanism, and each 10-point reduction in the score triggers a first-level warning.

[0010] Further, the working process of the legal clause matching engine in step S5 includes: S51. Construct a legal knowledge graph, parse the anti-price discrimination related legal provisions into structured (illegal act, recognition elements, penalty measures) triples, and map them to a high-dimensional feature space; S52. Calculate the matching degree between the real-time scenario feature vector and the legal provision feature vector, focusing on matching the three core dimensions of price difference rationality, user identity relevance, and subjective intention determination, and set dynamic weight coefficients for each dimension; S53. Generate a legal basis report. When the matching degree exceeds 0.75, automatically associate the case precedents of the same type in the most recent three years, and mark the median penalty amount and common defense reasons.

[0011] Further, the calculation method of the user price sensitivity index further includes: S311. Analyze the user's price comparison behavior characteristics, count the stay time and jump times of the user on the price comparison page, and generate a price attention index; S312. Calculate the coupon usage efficiency. By comparing the time difference after the coupon is received and the proportion of the order amount increase, evaluate the user's response speed to price incentives; S313. Construct an alternative product selection model, record the product switching frequency of the user in different price ranges, and when the frequency exceeds twice the average level of similar users, it is determined as a high price sensitivity feature.

[0012] Further, the calculation method of the supply and demand index deviation degree includes: S421. Collect real-time market data, including the median price of competing product platforms, inventory turnover rate, and search popularity index; S422. Calculate the supply-demand balance coefficient: ;

[0013] Wherein, is the average price of competing products, is the real-time search volume, is the current inventory, is the supply-demand balance coefficient, is the average search volume, is the price of the product on this platform, is the average inventory; S423. Normalize the supply-demand balance coefficient to generate an impact factor in the range of [0,1].

[0014] Furthermore, the construction method of the legal provision feature vector includes: S511. Parse legal provisions using natural language processing technology, and extract key elements of the constitutive elements of illegal acts, including the actor, subjective aspect, objective act, and damage result; S512. Map each element to a 64-dimensional semantic vector space, and calculate the correlation weight between elements through an attention mechanism; S513. When generating the provision feature vector, discretize and encode the penalty amount interval, and convert the limitation period of prosecution into an attenuation coefficient.

[0015] Furthermore, the multi-level intervention strategy in step S6 includes: S61. Under the high-risk warning level, execute simultaneously: Freeze the price adjustment interface of suspected abnormal products in real time Show the price comparison information flow of the product on different user terminals to users Automatically generate an electronic evidence collection package containing the time stamp, price comparison chart, and legal basis; S62. Under the medium-risk warning level, trigger: Extend the price update review period to at least 2 hours Send a secondary confirmation request to the operation staff, and they are required to provide an explanation of the reasonableness of the price adjustment; S63. Under the low-risk warning level, execute: Record the abnormal transaction log and mark the suspected feature label Start the tracking and monitoring of similar transaction behaviors in the next 7 days.

[0016] Furthermore, it also includes a model iteration and optimization mechanism: S71. Collect user feedback data and subsequent transaction behavior data after the warning is triggered, and construct a feedback training set; S72. Optimize the parameters of the price rationality scoring model, and focus on adjusting the proportional relationship between the supply and demand index influencing factor and the user group weight coefficient. The optimized objective function is: ;

[0017] where 、 is the balance coefficient, represents the mean absolute error, is the warning distribution KL divergence, is the model prediction value, is the true value, is the model prediction distribution, is the true distribution.

[0018] According to the above embodiments of the present invention, it has at least the following beneficial effects: The present invention can achieve accurate identification and real-time warning of abnormal price behaviors. By collecting user multi-dimensional behavior data in real time and performing spatio-temporal correlation processing, combined with the differential weighted fusion of user price sensitivity grading, it can accurately capture abnormal transaction characteristics and generate a classification feature matrix with high discrimination. On this basis, calculate the dynamic deviation index and generate a comprehensive price rationality score. When the score is lower than the preset risk level, trigger the warning mechanism, so as to timely discover potential abnormal price behaviors, provide a more transparent consumption environment for users, help them make more informed consumption decisions, and at the same time provide technical support for enterprises to assist them in optimizing pricing strategies. In addition, the present invention can also provide risk warnings for warning scenarios. Input the warning scenario feature vector into the legal clause matching engine, and generate a warning report containing relevant legal clause basis and reasonable explanations through legal text semantic analysis and scenario feature vector similarity calculation. This not only helps users safeguard their legitimate rights and interests when encountering abnormal price behaviors, but also provides clear compliance guidance for enterprises, improves their operation efficiency, and promotes the rationality and fairness of pricing. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By referring to the accompanying drawings and reading the following detailed description, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, wherein: Figure 1 is a flowchart of a method for identification warning and legal reminder based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only to enable those skilled in the art to better understand and then implement the present invention, and do not limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to fully convey the scope of the present invention to those skilled in the art.

[0021] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, equipment, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0022] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0023] The following refers to Figure 1 , Figure 1 is a schematic flowchart of a method for identification warning and legal prompt based on big data provided for an embodiment of the present invention. As Figure 1 shown, a method for identification warning and legal prompt based on big data includes: S1. Real-time collect multi-dimensional user behavior data, including user identity characteristic data, historical transaction price sequences, and real-time environmental variable data, where the user identity characteristic data at least includes device fingerprint characteristics, membership level labels, and consumption ability evaluation values, and the consumption ability evaluation value is obtained by weighted calculation of the user's historical order amount and consumption frequency; S2. Perform spatio-temporal association processing on the multi-dimensional behavior data, and generate a spatio-temporally aligned feature tensor through abnormal transaction time point detection and cross-platform user identity mapping, where the cross-platform user identity mapping is based on double verification of device fingerprint similarity and consumption behavior consistency; S3. Based on the user price sensitivity classification result, perform differential weighted fusion on the feature tensor to generate a user group classification feature matrix, where the historical transaction price fluctuation threshold of the high price sensitivity user group is reduced by 20%-30%; S4. Calculate the dynamic deviation index of the real-time transaction price relative to the user's historical benchmark price, and fuse the real-time supply and demand index and the promotion intensity parameter to generate a comprehensive price rationality score, and trigger an early warning mechanism when the score is lower than the preset risk level; S5. Input the early warning scenario feature vector into the legal clause matching engine to generate an early warning report including the legal basis and penalty clauses of the anti-price discrimination law, and the legal clause matching engine is implemented through legal text semantic analysis and scene feature vector similarity calculation; S6. Implement a multi-level intervention strategy according to the early warning report level, including pushing real-time price comparison information to the user terminal, sending interface throttling instructions to the enterprise risk control system, and submitting a risk record data packet to the supervision platform.

[0024] It should be noted that the present invention proposes a method for identification early warning and legal prompt based on big data. The core of this method lies in constructing a comprehensive user behavior portrait by collecting real-time multi-dimensional user behavior data, including user identity characteristic data, historical transaction price sequences, and real-time environmental variable data. The user identity characteristic data at least includes device fingerprint characteristics, membership level labels, and consumption ability evaluation values. Among them, the consumption ability evaluation value is obtained by weighted calculation of the user's historical order amount and consumption frequency. This indicator can effectively reflect the user's consumption level and purchasing power. The collection of multi-dimensional behavior data provides a rich data basis for subsequent analysis and early warning, enabling the system to evaluate and judge the user's trading behavior from multiple perspectives.

[0025] Specifically, the device fingerprint characteristics in the user identity characteristic data refer to generating a unique identifier by collecting multi-dimensional information such as the hardware information, software configuration, and network environment of the user's device, which is used to identify the behavior of the same device on different platforms. The membership level label is a level divided according to indicators such as the user's consumption amount and consumption frequency, such as ordinary member, gold member, platinum member, etc. Different membership levels usually correspond to different rights and services. The calculation method of the consumption ability evaluation value is to perform weighted calculation on the user's historical order amount and consumption frequency, and the weight can be adjusted according to the actual situation. For example, for some users with high consumption frequency but low consumption amount per time, the weight of consumption frequency can be appropriately increased. The historical transaction price sequence refers to the amount records of all transactions of the user within a certain period of time. By analyzing these data, the user's price sensitivity and purchasing habits can be understood. The real-time environmental variable data includes the current market supply and demand situation, promotional activities, etc. These factors will all affect the user's transaction price.

[0026] Preferably, when collecting multi-dimensional user behavior data, the data sources and types can be further refined. For example, during the collection of device fingerprint features, the user's geographical location information can be combined, and by analyzing data such as the user device's IP address and GPS positioning, it can be determined whether the user conducts transactions within a specific area, thereby excluding some abnormal transaction behaviors. For the historical transaction price sequence, a time window can be set, such as the transaction records in the past year or half-year, so as to more accurately reflect the user's recent consumption behavior. In terms of the collection of real-time environmental variable data, in addition to market supply and demand conditions and promotional activities, seasonal factors, holiday effects, etc. can also be considered, as these factors may have a significant impact on commodity prices. In addition, in order to improve the accuracy and reliability of the data, data cleaning and preprocessing techniques can be adopted to remove outliers and noisy data to ensure the effectiveness of subsequent analysis.

[0027] In some embodiments, step S2 includes: S21. Detect abnormal transaction time points using a dynamic time window. When it is detected that the fluctuation range of the transaction price within the time window exceeds 3 times the standard deviation of the same period in history, compensate and correct the original timestamp, and the correction formula is: ;

[0028] where represents the transaction price at time t, represents the average price within the window period, is a dynamic compensation coefficient and , is the sign function; S22. Construct a cross-platform user identity mapping graph. By calculating the similarity of device fingerprint hash values and the matching degree of consumption behavior sequences, generate a user identity confidence score. When the confidence score is greater than 0.8, it is determined as the same user entity; S23. Reconstruct the spatially and temporally aligned data into a three-dimensional feature tensor, with dimensions including the number of users, the number of time slices, and the number of feature channels. The feature channels at least include price volatility, change in consumption frequency, and environmental variable impact factor.

[0029] It should be noted that when processing multi-dimensional behavioral data, the present invention adopts a spatio-temporal correlation processing method. Through abnormal transaction time point detection and cross-platform user identity mapping, a spatio-temporally aligned feature tensor is generated. Abnormal transaction time point detection refers to analyzing the fluctuations of transaction prices within a specific time window, identifying transaction time points that are significantly different from historical data, and thus determining whether there are abnormal transaction behaviors. Cross-platform user identity mapping is based on the dual verification of device fingerprint similarity and consumption behavior consistency, associating the behaviors of the same user on different platforms to ensure the accuracy and integrity of the data. This process can effectively integrate user data from different platforms and provide a more comprehensive perspective for subsequent analysis.

[0030] Specifically, for abnormal transaction time point detection, a dynamic time window method is adopted. By calculating the fluctuation amplitude of transaction prices within the time window, when the fluctuation exceeds 3 times the standard deviation of the same period in history, the original time stamp is compensated and corrected. The dynamic compensation coefficient is adjusted according to the size of the fluctuation amplitude, ranging from 5 milliseconds to 30 milliseconds. Cross-platform user identity mapping is achieved by constructing a user identity confidence score, which is calculated based on the similarity of device fingerprint hash values and the matching degree of consumption behavior sequences. When the confidence score is greater than 0.8, it is determined as the same user entity. The spatio-temporally aligned data is reconstructed into a three-dimensional feature tensor, whose dimensions include the number of users, the number of time slices, and the number of feature channels. Among them, the feature channels at least include price volatility, change in consumption frequency, and environmental variable impact factor.

[0031] Preferably, in abnormal transaction time point detection, the size of the time window and the range of the dynamic compensation coefficient can be adjusted according to the transaction characteristics of different commodities. For example, for commodities with high-frequency trading, the time window can be set shorter, and the dynamic compensation coefficient can be appropriately reduced to improve the detection sensitivity. In cross-platform user identity mapping, in addition to the verification of device fingerprint and consumption behavior consistency, social network information of users can also be introduced as an auxiliary verification means to further improve the accuracy of user identity recognition. In addition, for the construction of the feature tensor, more feature channels can be added, such as user evaluation data, browsing behavior data, etc., to enrich the data dimensions and provide more comprehensive information for subsequent analysis.

[0032] In some embodiments, the differential weighted fusion in step S3 includes: S31. Calculate the user price sensitivity index, which is comprehensively evaluated by analyzing three dimensions of price comparison response time, coupon usage efficiency, and alternative commodity selection tendency in the user's historical orders, where the weight ratio of the alternative commodity selection tendency is not less than 40%; S32. Build a user group classification model, divide the users with the top 10% price sensitivity index into the high-sensitivity group, the middle 30% into the medium-sensitivity group, and the remaining into the low-sensitivity group, and set different price fluctuation tolerance coefficients for different groups; S33. Perform group weighted pooling processing on the feature tensor, increase the feature channel weights of the high-sensitivity group by 1.5 - 2 times, and reduce the weights of the low-sensitivity group to 0.7 - 0.9 times to generate a classification feature matrix with group discrimination.

[0033] It should be noted that when processing the feature tensor in the present invention, a differential weighted fusion mechanism based on user price sensitivity classification is introduced. User price sensitivity classification divides users into different groups according to their response degrees to price changes, thereby providing a basis for subsequent weighted fusion. Differential weighted fusion assigns different weights to each feature channel in the feature tensor according to the price sensitivity of the user group to generate a classification feature matrix with group discrimination. This process can more accurately reflect the transaction behavior characteristics of different user groups and provide more accurate data support for subsequent price rationality evaluation.

[0034] Specifically, the calculation of the user price sensitivity index involves multiple dimensions, including price comparison response time, coupon usage efficiency, and alternative product selection tendency. The price comparison response time refers to the stay duration and jump times of the user viewing the price comparison page, reflecting the user's attention to price. The coupon usage efficiency is evaluated by comparing the time difference after receiving the coupon and the proportion of order amount increase to measure the user's response speed to price incentives. The alternative product selection tendency refers to the switching frequency of the user among products in different price ranges. When the frequency exceeds twice the average level of similar users, it is determined as a high price sensitivity feature. When building the user group classification model, the users with the top 10% price sensitivity index are divided into the high-sensitivity group, the middle 30% into the medium-sensitivity group, and the remaining into the low-sensitivity group. Different groups are set with different price fluctuation tolerance coefficients, and the historical transaction price fluctuation threshold of the high-sensitivity group is reduced by 20% - 30%.

[0035] Preferably, when calculating the user price sensitivity index, the weight allocation of each dimension can be further refined. For example, for the dimension of alternative product selection tendency, considering its key role in reflecting user price sensitivity, its weight ratio can be set to no less than 40%. When constructing the user group classification model, the division ratio of the user group can be adjusted according to the actual application scenario. For example, in some markets with fierce price competition, the proportion of highly sensitive groups can be appropriately increased to more accurately capture the behavioral characteristics of price-sensitive users. In addition, for the group weighted pooling process of the feature tensor, a dynamic adjustment mechanism can be introduced to dynamically adjust the feature channel weights of the highly sensitive group and the low-sensitive group according to real-time market data and user behavior changes, further improving the accuracy and timeliness of the classification feature matrix.

[0036] In some embodiments, step S4 includes: S41. Generate the user historical benchmark price curve, and fuse the user's past K transaction prices using the exponential decay weighting method, where the weight ratio of the transaction prices in the most recent 30 days is not less than 60%; S42. Calculate the dynamic price deviation index, generate a normalized deviation score by comparing the absolute difference between the real-time transaction price and the benchmark price and superimposing the influence factor of the real-time supply and demand index deviation. The specific formula is: ;

[0037] where, is the benchmark price, represents the deviation degree of the real-time supply and demand index from the historical benchmark value, is the supply and demand index benchmark value; S43. Construct a price rationality scoring model. When the deviation index exceeds the tolerance threshold of the user group, start the price rationality downgrading mechanism, and each 10-point reduction in the score triggers a first-level warning.

[0038] It should be noted that when evaluating the rationality of the real-time transaction price in the present invention, a comprehensive price rationality score is generated by calculating the dynamic price deviation index and combining the real-time supply and demand index and the promotion intensity parameter. The dynamic price deviation index is an indicator that measures the difference between the real-time transaction price and the user's historical benchmark price, while the real-time supply and demand index and the promotion intensity parameter reflect the impact of the current market environment on the price. This comprehensive scoring mechanism can more comprehensively evaluate the rationality of the transaction price, thereby effectively identifying potential differential pricing behaviors.

[0039] Specifically, the user's historical benchmark price curve is generated by fusing the user's past K transaction prices through the exponential decay weighting method, where the weight ratio of the transaction prices in the most recent 30 days is not less than 60% to ensure that the benchmark price can reflect the user's recent consumption situation. The dynamic price deviation index is generated by comparing the absolute difference between the real-time transaction price and the benchmark price and superimposing the influencing factor of the real-time supply-demand index deviation. The real-time supply-demand index deviation degree is obtained by collecting real-time market data, including the median price of competing product platforms, inventory turnover rate, search popularity index, etc., calculating the supply-demand balance coefficient and performing normalization processing. The price rationality scoring model activates the price rationality downgrading mechanism according to whether the deviation index exceeds the tolerance threshold of the user's group, and triggers an alarm when the score is lower than the preset risk level.

[0040] Preferably, when generating the user's historical benchmark price curve, the weight distribution of the past K transaction prices can be dynamically adjusted according to the consumption cycle and market volatility characteristics of different commodities. For example, for seasonal commodities, the weight of the transaction prices within the most recent quarter can be increased, while for daily consumer goods, more emphasis can be placed on the transaction prices in the most recent month. When calculating the dynamic price deviation index, more market factors can be introduced, such as the type and intensity of promotional activities, to further refine the calculation method of the real-time supply-demand index deviation degree. For example, different influencing factor weights can be set for time-limited discounts and full reduction activities to more accurately reflect the impact of promotional activities on price rationality. In addition, in the price rationality scoring model, the tolerance threshold can be dynamically adjusted according to the price sensitivity of the user group. For high-price-sensitivity user groups, a more stringent threshold can be set to improve the accuracy of the alarm.

[0041] In some embodiments, the workflow of the legal clause matching engine in step S5 includes: S51. Construct a legal knowledge graph, parse the anti-price discrimination related legal provisions into structured (illegal act, determination elements, penalty measures) triples, and map them to a high-dimensional feature space; S52. Calculate the matching degree between the real-time scenario feature vector and the legal provision feature vector, focusing on matching the three core dimensions of price difference rationality, user identity relevance, and subjective intention determination, and set dynamic weight coefficients for each dimension; S53. Generate a legal basis report. When the matching degree exceeds 0.75, automatically associate the case precedents of the same type in the most recent three years, and mark the median penalty amount and common defense reasons.

[0042] It should be noted that in the present invention, the early warning scenario feature vector is matched with legal provisions through a legal clause matching engine to generate an early warning report containing the legal basis of anti-price discrimination laws and penalty clauses. The legal clause matching engine uses a legal knowledge graph and the calculation of the similarity of scenario feature vectors to achieve the legal assessment of differential pricing behavior. The legal knowledge graph analyzes relevant anti-price discrimination legal provisions into structured triples (illegal acts, elements for determination, penalty measures) and maps them to a high-dimensional feature space for matching with real-time scenario feature vectors. This mechanism can provide clear legal basis for regulatory authorities and users and enhance the deterrence against differential pricing behavior.

[0043] Specifically, the working process of the legal clause matching engine includes three main steps. First, when constructing the legal knowledge graph, natural language processing technology is used to analyze legal provisions, extract key elements of the constitutive elements of illegal acts, such as the subject of the act, subjective aspects, objective acts, and damage results, and map these elements to a 64-dimensional semantic vector space. Second, calculate the matching degree between the real-time scenario feature vector and the legal provision feature vector, focusing on matching the three core dimensions of the rationality of price difference, the relevance of user identity, and the determination of subjective intention, and set dynamic weight coefficients for each dimension. Finally, when the matching degree exceeds 0.75, a legal basis report is automatically generated, and case precedents of the same type in the recent three years are associated, marking the median penalty amount and common defenses. This process ensures the accuracy and timeliness of the legal assessment.

[0044] Preferably, when constructing the legal knowledge graph, the parsing process of legal provisions can be further refined. For example, for the penalty amount range, a discretization coding method can be used to convert the continuous amount range into discrete categories for subsequent matching and analysis. At the same time, for the limitation of action period, it can be converted into an attenuation coefficient to reflect the impact of time on legal effect. When calculating the matching degree, the weight coefficients of the three core dimensions can be dynamically adjusted according to the complexity of different scenarios. For example, in transactions involving high-value goods, the weight of the rationality of price difference dimension can be increased, while in scenarios involving user privacy data, the weight of the relevance of user identity dimension can be increased. In addition, to improve the practicality of the legal basis report, a detailed explanation of relevant legal provisions and description of applicable conditions can be added to the report to help users and regulatory authorities better understand and apply the legal basis.

[0045] In some embodiments, the calculation method of the user price sensitivity index further includes: S311. Analyze the user price comparison behavior characteristics, count the stay duration and jump times of the user viewing the price comparison page, and generate a price attention index; S312. Calculate the coupon usage efficiency, and evaluate the response speed of users to price incentives by comparing the time difference between receiving and using the coupon and the increase ratio of the order amount. S313. Build a substitute product selection model, record the product switching frequency of users in different price ranges, and when the frequency exceeds twice the average level of users of the same type, it is determined as a high price sensitivity feature.

[0046] It should be noted that when calculating the user price sensitivity index of the present invention, the analysis methods of user price comparison behavior characteristics, coupon usage efficiency, and substitute product selection tendency are further refined. User price comparison behavior characteristics refer to the behavior patterns of users viewing price comparison pages during the shopping process, including the stay duration and the number of jumps, so as to measure the degree of users' attention to prices. The coupon usage efficiency is evaluated by analyzing the usage situation of users after receiving the coupon to assess the response speed of users to price incentives. The substitute product selection tendency refers to the frequency of users switching products in different price ranges, and this indicator can effectively reflect the sensitivity of users to price changes. By comprehensively evaluating these three dimensions, the user price sensitivity index can be calculated more accurately.

[0047] Specifically, the analysis of user price comparison behavior characteristics includes counting the stay duration and the number of jumps of users viewing the price comparison page to generate a price attention degree indicator. For example, if a user stays on the price comparison page for a long time and frequently jumps to other product pages, it indicates that the user is more concerned about prices, and its price attention degree indicator will increase accordingly. The calculation of the coupon usage efficiency is evaluated by comparing the time difference between receiving and using the coupon and the increase ratio of the order amount. For example, if a user uses the coupon shortly after receiving it and the order amount has a significant increase, it means that the user has a fast response speed to price incentives and a high coupon usage efficiency. The analysis of the substitute product selection tendency is to record the product switching frequency of users in different price ranges, and when the frequency exceeds twice the average level of users of the same type, it is determined as a high price sensitivity feature. For example, within the same price range, if a user frequently switches different brands or models of products, it indicates that the user is more sensitive to price changes.

[0048] Preferably, when analyzing the behavioral characteristics of users' price comparison, the weight settings of the stay duration and the number of jumps can be further refined. For example, according to different product categories and market environments, dynamically adjust the weight ratio of the stay duration and the number of jumps. For high-value products, the weight of the stay duration can be increased because users usually spend more time comparing when purchasing such products. When calculating the coupon usage efficiency, more parameters can be introduced, such as the denomination of the coupon, the usage scenario, etc., to more comprehensively evaluate the user's response to price incentives. For example, for large-denomination coupons, users may be more inclined to use them on high-value products, so different usage efficiency evaluation criteria can be set for coupons of different denominations.

[0049] Furthermore, in the analysis of the alternative product selection tendency, the user purchase conversion rate can be introduced as an auxiliary indicator to more accurately reflect the user's price sensitivity. For example, if the user finally completes the purchase after frequently switching products, it indicates that the user is highly sensitive to price changes, but the purchase conversion rate is also high, which means that the user is looking for products with higher cost performance.

[0050] In some embodiments, the calculation method of the supply-demand index deviation degree includes: S421. Collect real-time market data, including the median price of competing product platforms, inventory turnover rate, and search popularity index; S422. Calculate the supply-demand balance coefficient: ;

[0051] Wherein, is the average price of competing products, which refers to the median price of other platforms in the same or similar product categories, is the real-time search volume, that is, the search volume of the current product, reflecting the demand heat of the market for this product, is the current inventory, representing the inventory quantity of the current product, reflecting the supply situation of the market, is the supply-demand balance coefficient, is the average search volume, which is the average value of historical search volumes, used to calculate the relative change of the search volume to measure the fluctuation of market demand, is the price of the product on this platform, is the average inventory, which is the historical average inventory, used as a comparison benchmark to evaluate the difference between the current inventory level and the historical average level; S423. Normalize the supply-demand balance coefficient to generate an influence factor in the range of [0,1].

[0052] It should be noted that when calculating the deviation degree of the supply-demand index in the present invention, the influencing factors are generated by collecting real-time market data and calculating the supply-demand balance coefficient. The deviation degree of the supply-demand index reflects the difference between the current market supply-demand relationship and the historical average level, and is an important reference index for evaluating price rationality. The real-time market data includes the median price of competing product platforms, inventory turnover rate, and search popularity index, etc., which can comprehensively reflect market dynamics. The supply-demand balance coefficient is obtained by weighted calculation of these data, and the finally generated normalized influencing factor is used to adjust the relevant parameters in the price rationality scoring model.

[0053] Specifically, when calculating the deviation degree of the supply-demand index, first collect real-time market data, including the median price of competing product platforms, inventory turnover rate, and search popularity index. The median price of competing product platforms refers to the median price of other platforms in the same or similar product categories; the inventory turnover rate is the ratio of the current inventory quantity to the historical average inventory quantity; the search popularity index is the ratio of the current search volume of the product to the historical average search volume. The supply-demand balance coefficient is obtained by weighted calculation of these data, and the weight of each parameter can be adjusted according to the market characteristics and historical data of the product. For example, for seasonal products, the weight of the search popularity index can be appropriately increased because it is greatly affected by seasonality. Finally, the supply-demand balance coefficient is normalized to generate an influencing factor within the range of [0,1], which is used to adjust the relevant parameters in the price rationality scoring model.

[0054] Preferably, when collecting real-time market data, the data source and collection frequency can be further refined. For example, for the median price of competing product platforms, the data collection frequency can be increased from once per hour to once every 15 minutes to more timely reflect market dynamics. When calculating the supply-demand balance coefficient, the weights of each parameter can be dynamically adjusted according to the market fluctuation characteristics of different products. For example, for products with high demand and low inventory, the weight of the inventory turnover rate can be increased to more accurately reflect the market supply-demand relationship. In addition, different normalization methods can be adopted, such as maximum-minimum normalization or Z-score normalization, and the specific method can be selected according to the distribution characteristics of the data. For example, when the data distribution is relatively uniform, maximum-minimum normalization can be adopted; when there are outliers in the data distribution, Z-score normalization can be adopted to improve the accuracy and stability of the influencing factor.

[0055] In some embodiments, the construction method of the legal provision feature vector includes: S511. Parse the legal provisions using natural language processing technology, and extract the key elements of the constitutive elements of illegal acts, including the actor, subjective aspect, objective act, and damage result; S512. Map each element to a 64-dimensional semantic vector space, and calculate the correlation weight between elements through an attention mechanism; In S513, when generating the feature vector of the legal provision, the penalty amount range is discretely encoded, and the limitation period of litigation is converted into an attenuation coefficient.

[0056] It should be noted that when constructing the feature vector of the legal provision in the present invention, natural language processing technology is used to parse the legal provision, extract the key elements of the constitutive elements of illegal acts, and map these elements into a high-dimensional semantic vector space. The construction of the feature vector of the legal provision is the core part of the legal provision matching engine. It provides a basis for subsequent matching and evaluation by converting the semantic information of the legal provision into a computable vector form. This process not only improves the operability of the legal provision but also enhances the accuracy and efficiency of legal evaluation.

[0057] Specifically, when constructing the feature vector of the legal provision, first, natural language processing technology is used to parse the legal provision, and the key elements of the constitutive elements of illegal acts are extracted, including the subject of the act, the subjective aspect, the objective act, and the harmful consequences. The subject of the act refers to the individual or organization that commits the illegal act; the subjective aspect refers to the subjective intention of the actor, such as intention or negligence; the objective act refers to the specific act implemented, such as price discrimination; the harmful consequences refer to the specific consequences caused by the act, such as the infringement of consumers' rights and interests. These elements are mapped into a 64-dimensional semantic vector space, and the correlation weights between the elements are calculated through the attention mechanism. When generating the feature vector of the legal provision, the penalty amount range is discretely encoded, and the limitation period of litigation is converted into an attenuation coefficient. Discretely encoding is to divide the continuous penalty amount range into several discrete categories for subsequent matching and analysis; the attenuation coefficient quantifies the timeliness of the legal provision according to the length of the limitation period of litigation.

[0058] Preferably, when extracting the key elements of the legal provision, the application of natural language processing technology can be further refined. For example, pre-trained language models in deep learning (such as BERT or Transformer) are used to improve the accuracy and efficiency of text parsing. When mapping into the semantic vector space, the dimension of the vector space can be dynamically adjusted according to the complexity of the legal provision. For example, for complex provisions involving multiple legal fields, the dimension of the vector space can be increased to better capture semantic information. When discretely encoding the penalty amount range, the amount range can be flexibly divided according to the actual situation of different legal fields or specific cases. For example, in cases involving high-value goods, the penalty amount range can be divided into finer categories to more accurately reflect the penalty intensity. In addition, when converting the limitation period of litigation into an attenuation coefficient, a time decay function, such as exponential decay or logarithmic decay, can be introduced to more accurately reflect the impact of time on the legal effect.

[0059] In some embodiments, the multi-level intervention strategy in step S6 includes: S61. Under the high-risk warning level, the following operations are executed simultaneously: Real-time freeze the price adjustment interface of suspected abnormal goods Show the price comparison information flow of the goods on different user terminals to the user Automatically generate an electronic forensics package including timestamp, price comparison chart, and legal basis; S62. Under the medium-risk warning level, the following are triggered: Extend the price update review period to at least 2 hours Send a secondary confirmation request to the operator, who is required to provide an explanation for the reasonableness of the price adjustment; S63. Under the low-risk warning level, the following are executed: Record the abnormal transaction log and mark the suspected feature labels Start the tracking and monitoring of similar transaction behaviors in the next 7 days.

[0060] It should be noted that when implementing the multi-level intervention strategy in the present invention, different intervention measures are taken according to the level of the warning report. The multi-level intervention strategy aims to adopt corresponding measures to deal with potential differential pricing behaviors through warning signals of different levels. Under the high-risk warning level, measures such as freezing the price adjustment interface, showing the price comparison information flow, and generating an electronic forensics package are executed simultaneously; under the medium-risk warning level, extending the price update review period and sending a secondary confirmation request are triggered; under the low-risk warning level, recording the abnormal transaction log and starting the tracking and monitoring are carried out. This hierarchical intervention mechanism can effectively balance the regulatory intensity and market flexibility, ensuring the normal operation of the market while protecting the rights and interests of consumers.

[0061] Specifically, the implementation of the multi-level intervention strategy involves multiple specific operation steps. Under the high-risk warning level, the system will real-time freeze the price adjustment interface of suspected abnormal goods to prevent merchants from further adjusting prices. At the same time, show the price comparison information flow of the goods on different user terminals to the user to help the user understand the price difference. In addition, the system will automatically generate an electronic forensics package including timestamp, price comparison chart, and legal basis to provide evidence support when needed. Under the medium-risk warning level, the system will extend the price update review period to at least 2 hours, which provides sufficient time for supervisors to review the reasonableness of the price adjustment. At the same time, send a secondary confirmation request to the operator, asking for an explanation of the reasonableness of the price adjustment. Under the low-risk warning level, the system will record the abnormal transaction log and mark the suspected feature labels, and at the same time start the tracking and monitoring of similar transaction behaviors in the next 7 days to continuously observe the market dynamics.

[0062] Preferably, when implementing the multi-level intervention strategy, the specific operations of each warning level can be further refined. For example, under the high-risk warning level, in addition to the above measures, real-time warnings to merchants can be added to remind them of the possible legal risks in price adjustments. Under the medium-risk warning level, an artificial intelligence-assisted audit mechanism can be introduced to quickly analyze the rationality explanations provided by operators and improve the audit efficiency. Under the low-risk warning level, an automatic marking function for abnormal transactions can be added to facilitate quick positioning of problems during subsequent analysis. In addition, the thresholds of the warning levels can be dynamically adjusted according to market dynamics and user feedback. For example, when the market fluctuates greatly, the warning threshold can be appropriately increased to reduce false alarms; when the market is stable, the threshold can be lowered to improve the sensitivity of the warning.

[0063] In some embodiments, it further includes a model iteration and optimization mechanism: S71. Collect user feedback data and subsequent transaction behavior data after the warning is triggered to construct a feedback training set; S72. Optimize the parameters of the price rationality scoring model, focusing on adjusting the proportional relationship between the supply and demand index influence factor and the user group weight coefficient. The optimization objective function is: ;

[0064] Among them, , is a balance coefficient used to adjust the weights of the mean absolute error (MAE) and KL divergence in the optimization objective function to achieve a balance between the accuracy and generalization ability of the model; represents the mean absolute error, which is used to measure the average absolute difference between the model prediction value and the true value; is the KL divergence of the warning distribution, which is used to measure the similarity between the model prediction distribution P and the true distribution Q in the optimization objective function. The smaller the KL divergence value, the more similar the two distributions are, that is, the closer the model prediction distribution is to the true distribution, and the stronger the generalization ability of the model; is the model prediction value. The model prediction value is the prediction result of the price rationality scoring model for the rationality score of the transaction price. By optimizing the model parameters, make as close as possible to the true price rationality score ; is the true value, which refers to the actual price rationality score and is the benchmark value used for comparison during model training and optimization, reflecting the rationality degree of the transaction price in the actual situation; is the model prediction distribution, which is the probability distribution formed by the prediction results of the price rationality scoring model for the rationality scoring of transaction prices, reflecting the prediction possibilities of the model for different scoring situations; is the true distribution, referring to the actual probability distribution of the price rationality scoring of transaction prices, representing the true situation of the price rationality scoring in the market. The model optimizes the parameters to make the prediction distribution as close as possible to the true distribution .

[0065] It should be noted that the present invention introduces a model iteration and optimization mechanism. By collecting user feedback data and subsequent transaction behavior data after the warning is triggered, a feedback training set is constructed, and then the parameters of the price rationality scoring model are optimized. This mechanism aims to improve the recognition accuracy of the model for differential pricing behavior and the reliability of warnings through continuous learning and optimization. The core of the model iteration and optimization mechanism lies in dynamically adjusting the proportional relationship between the supply and demand index influence factor and the user group weight coefficient to better adapt to the dynamic characteristics of market changes and user behavior.

[0066] Specifically, the model iteration and optimization mechanism involves two main steps. First, collect user feedback data after the warning is triggered. These data include the user's response to the warning information, subsequent transaction behavior, and evaluation of price adjustments, etc. These data are used to construct a feedback training set to provide a basis for model optimization. Second, optimize the parameters of the price rationality scoring model, focusing on adjusting the proportional relationship between the supply and demand index influence factor and the user group weight coefficient. The optimization objective function is achieved by minimizing the weighted sum of the mean absolute error (MAE) and the KL divergence. The mean absolute error measures the difference between the model prediction value and the actual value, while the KL divergence is used to measure the similarity between the model prediction distribution and the true distribution. By adjusting the balance coefficient, a balance can be achieved between the accuracy and generalization ability of the model.

[0067] Preferably, when constructing the feedback training set, the collection method of user feedback data can be further refined. For example, the satisfaction of users with the warning information and subsequent behaviors can be collected through user surveys, online feedback forms, or feedback functions in mobile applications. When optimizing the price rationality scoring model, the supply and demand index influence factor and the user group weight coefficient can be dynamically adjusted according to the characteristics of different market environments and user groups. For example, in a market with large demand fluctuations, the weight of the supply and demand index influence factor can be increased; in a market with large differences among user groups, the weight of the user group weight coefficient can be increased. In addition, more optimization algorithms, such as genetic algorithms or Bayesian optimization, can be introduced to improve the efficiency and effect of model parameter optimization. These algorithms can automatically adjust the parameters according to the performance of the model, further improving the accuracy and reliability of the model.

[0068] The above-mentioned various embodiments of the present invention have the following beneficial effects: The present invention can achieve accurate identification and real-time early warning of abnormal price behaviors. By collecting users' multi-dimensional behavior data in real time and performing spatio-temporal correlation processing, combined with the differential weighted fusion of users' price sensitivity grading, it can accurately capture abnormal transaction characteristics and generate a classification feature matrix with high discrimination. On this basis, a dynamic deviation index is calculated and a comprehensive price rationality score is generated. When the score is lower than the preset risk level, the early warning mechanism is triggered, so as to timely discover potential abnormal price behaviors, provide a more transparent consumption environment for users, help them make more informed consumption decisions, and at the same time provide technical support for enterprises to assist them in optimizing pricing strategies.

[0069] In addition, the present invention can also provide legal basis and relevant clause prompts for the early warning scenario. Input the feature vector of the early warning scenario into the legal clause matching engine. Through semantic analysis of legal provisions and similarity calculation of scenario feature vectors, an early warning report containing relevant legal clause basis and reasonable explanations is generated. This not only helps users safeguard their legitimate rights and interests when encountering abnormal price behaviors, but also provides clear compliance guidance for enterprises, improves their operation efficiency, and promotes the rationality and fairness of market pricing.

[0070] The present invention can also adopt a multi-level intervention strategy, take corresponding measures according to different early warning levels, and effectively curb the occurrence of abnormal price behaviors. At the high-risk early warning level, the price adjustment interface of suspected abnormal goods can be frozen in real time, the price comparison information flow of the goods on different user terminals can be displayed to users, and an electronic evidence collection package containing time stamps, price comparison charts, and legal basis can be automatically generated; at the medium-risk early warning level, the price update review period can be extended to at least 2 hours, and a secondary confirmation request can be sent to the operation personnel, who are required to provide an explanation of the rationality of price adjustment; at the low-risk early warning level, abnormal transaction logs can be recorded and suspected feature tags can be marked, and the subsequent 7-day tracking and monitoring of similar transaction behaviors can be started.

[0071] In addition, the present invention can also optimize the model through an iterative optimization mechanism, collect users' feedback data and subsequent transaction behavior data after the early warning is triggered, construct a feedback training set, optimize the parameters of the price rationality scoring model, and focus on adjusting the proportional relationship between the supply and demand index influencing factor and the user group weight coefficient to further improve the accuracy and reliability of the model.

[0072] Further, the storage medium according to the embodiment of the present application stores program instructions capable of implementing all of the above methods. Among them, the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.

[0073] The above description is only some preferred embodiments of the present invention and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for identifying, warning and legal reminder based on big data, characterized in that: The following steps are involved: S1. Real-time collection of multi-dimensional user behavior data, including user identity feature data, historical transaction price series, and real-time environmental variable data, wherein the user identity feature data at least includes device fingerprint features, member level labels, and consumption capacity evaluation values, and the consumption capacity evaluation values ​​are obtained by weighted calculation of the user's historical order amount and consumption frequency; S2. Performing spatiotemporal correlation processing on the multi-dimensional behavior data, generating a spatiotemporal aligned feature tensor through abnormal transaction time point detection and cross-platform user identity mapping, wherein the cross-platform user identity mapping is based on dual verification of device fingerprint similarity and consumption behavior consistency; S3. Based on the user price sensitivity classification results, perform differentiated weighted fusion on the feature tensor to generate a user group classification feature matrix, wherein the historical transaction price fluctuation threshold of the high price sensitivity user group is reduced by 20%-30%; S4. Calculate the dynamic deviation index of the real-time transaction price relative to the user's historical benchmark price, and integrate the real-time supply and demand index and the promotion intensity parameter to generate a comprehensive score for price rationality. When the score is lower than the preset risk level, trigger the early warning mechanism; S5. Input the warning scenario feature vector into the legal clause matching engine to generate a warning report containing the anti-price discrimination legal basis and penalty clauses, wherein the legal clause matching engine is implemented by legal clause semantic analysis and scenario feature vector similarity calculation; S6. Implement multi-level intervention strategies based on the warning report level, including pushing real-time price comparison information to user terminals, sending interface flow limiting instructions to the enterprise risk control system, and submitting risk filing data packages to the regulatory platform.

2. The method according to claim 1, characterized in that Step S2 includes: S21. Use dynamic time window to detect abnormal trading time points. When it is detected that the trading price is within the time window, When the fluctuation range exceeds 3 times the standard deviation of the historical period, the original timestamp is compensated and corrected. The correction formula is: ； in, represents the transaction price at time t, represents the average price during the window period, is the dynamic compensation coefficient, is the symbolic function, is the corrected timestamp, is the original timestamp; S22. Build a cross-platform user identity mapping map, generate a user identity confidence score by calculating the similarity of the device fingerprint hash value and the matching degree of the consumption behavior sequence, and determine that it is the same user entity when the confidence score is greater than 0.8; S23. Reconstruct the spatiotemporally aligned data into a three-dimensional feature tensor, the dimensions of which include the number of users, the number of time slices, and the number of feature channels, where the feature channels at least include price volatility, consumption frequency change, and environmental variable influencing factors.

3. The method according to claim 1, characterized in that The differential weighted fusion in step S3 includes: S31. Calculate the user price sensitivity index and conduct a comprehensive evaluation by analyzing the price comparison response time, coupon usage efficiency, and alternative product selection tendency in the user's historical orders; S32. Build a user group classification model, divide the top 10% of users in price sensitivity index into a high-sensitivity group, the middle 30% into a medium-sensitivity group, and the rest into a low-sensitivity group, and set differentiated price fluctuation tolerance coefficients for different groups; S33. Perform group weighted pooling processing on the feature tensor to increase the feature channel weights of the highly sensitive group and generate a classification feature matrix with group discrimination.

4. The method according to claim 1, characterized in that Step S4 includes: S41. Generate a user's historical benchmark price curve, using the exponential decay weighting method to integrate the user's past K transaction prices, where the weight of the transaction prices in the last 30 days accounts for no less than 60%; S42. Calculate the dynamic price deviation index by comparing the absolute difference between the real-time transaction price and the benchmark price and superimposing the influencing factor of the real-time supply and demand index deviation to generate a normalized deviation score. The specific formula is: ； in, is the base price, Indicates the deviation between the real-time supply and demand index and the historical benchmark value. is the benchmark value of the supply and demand index, is the normalized deviation score, Real-time transaction price; S43. Construct a price rationality scoring model. When the deviation index exceeds the tolerance threshold of the user's group, activate the price rationality downgrade mechanism. Every 10-point drop in the score triggers a first-level warning.

5. The method according to claim 1, characterized in that: The legal terms matching engine workflow in step S5 includes: S51. Construct a legal knowledge graph, parse the relevant anti-price discrimination legal provisions into structured illegal behavior, identification requirements, and punishment measures triples, and map them to a high-dimensional feature space; S52, calculating the matching degree between the real-time scenario feature vector and the legal provision feature vector, focusing on matching the three core dimensions of price difference rationality, user identity relevance, and subjective intent determination, and setting a dynamic weight coefficient for each dimension; S53. Generate a legal basis report. When the matching degree exceeds 0.75, it will automatically link the case law of similar cases in the past three years, and mark the median penalty amount and common grounds for defense.

6. The method according to claim 3, characterized in that The method for calculating the user price sensitivity index also includes: S311, analyzing the characteristics of user price comparison behavior, counting the length of time users stay on the price comparison page and the number of jumps, and generating a price attention index; S312, calculating the coupon usage efficiency, by comparing the usage time difference after coupon collection and the increase ratio of order amount, evaluating the user's response speed to price incentives; S313. Construct an alternative product selection model to record the user's switching frequency of products in different price ranges. When the frequency is more than twice the average level of similar users, it is determined to be a high price sensitivity feature.

7. The method according to claim 4, characterized in that The calculation method of the supply and demand index deviation degree includes: S421. Collect real-time market data, including median prices on competing platforms, inventory turnover rate, and search popularity index; S422. Calculate the supply-demand balance coefficient: ； in, is the average price of competing products, is the real-time search volume. is the current inventory, is the supply-demand balance coefficient, is the mean search volume, For the price of products on this platform, is the average inventory; S423. Normalize the supply-demand balance coefficient to generate an impact factor in the interval [0,1].

8. The method according to claim 5, characterized in that The method for constructing the legal text feature vector includes: S511. Use natural language processing technology to parse legal provisions and extract key elements of the elements of illegal behavior, including the subject of the behavior, subjective aspects, objective behavior, and damage results; S512, mapping each element to a 64-dimensional semantic vector space, and calculating the association weights between the elements through an attention mechanism; S513. When generating the feature vector of the article, the penalty amount interval is discretized and encoded, and the limitation period is converted into an attenuation coefficient.

9. The method according to claim 1, characterized in that: The multi-level intervention strategy in step S6 includes: S61, at high risk warning level, execute simultaneously: Real-time freezing of the price adjustment interface for suspected abnormal products; Show users the price comparison information flow of the product on different user terminals; Automatically generate electronic evidence packages containing timestamps, price comparison charts, and legal basis; S62, medium-risk warning level, triggers: Extend the price update review period to a minimum of 2 hours; Send a second confirmation request to the operator, and provide a reasonable explanation for the price adjustment; S63, low risk warning level, execute: Record abnormal transaction logs and mark suspected feature labels; Initiate tracking and monitoring of similar transaction behaviors for the next 7 days.

10. The method according to claim 4, characterized in that It also includes a model iteration optimization mechanism: S71. Collect user feedback data and subsequent transaction behavior data after the warning is triggered, and construct a feedback training set; S72. Optimize the parameters of the price rationality scoring model, focus on adjusting the proportional relationship between the supply and demand index influencing factor and the user group weight coefficient, and optimize the objective function as follows: ； in, , is the balance coefficient, represents the mean absolute error, For the KL divergence of the warning distribution, is the model prediction value, is the true value, is the model prediction distribution, is the true distribution.

Images