An AI-based intelligent response system and method for platform risk events

Abstract

This invention discloses an intelligent response system and method for platform risk events based on artificial intelligence, relating to the field of risk detection technology. The invention collects product information published on e-commerce platforms and product review information from historical transactions, calculating the degree of difference between the two; it filters target products with excessive differences, and calculates a risk warning coefficient by combining information modification rate, complaint rate, and return rate; it uses the risk warning coefficient and the degree of difference as two-dimensional indicators to classify risk levels, and matches corresponding risk response strategies according to the risk levels. Simultaneously, it constructs a multi-dimensional evaluation system to calculate the response effect coefficient, dynamically adjusting the risk response strategy. This invention can achieve intelligent identification and dynamic response to risk events, protecting user rights and platform reputation.

Description

Technical Field

[0001] This invention relates to the field of risk detection technology, specifically to an intelligent response system and method for platform-based risk events based on artificial intelligence. Background Technology

[0002] E-commerce platforms refer to online trading platforms established on the internet, providing online transaction and payment functions for buyers and sellers. As e-commerce matures and its use becomes more widespread, risks are also becoming increasingly prominent. For the sustainable development of e-commerce platforms, in terms of risk event response, platforms often use manual sampling or keyword blocking methods to prevent false advertising. Manual sampling is not only inefficient but also inaccurate. Regarding keyword blocking, some unscrupulous merchants can circumvent the platform's keyword blocking rules through homophone substitution, synonym substitution, and word splitting / recombination, continuing to publish misleading product information. Therefore, there is an urgent need for an intelligent, multi-dimensional risk event response method to address these problems. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent response system and method for platform risk events based on artificial intelligence, so as to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent response method for platform risk events based on artificial intelligence, the intelligent response method comprising:

[0005] Step S1: Collect product information published on the e-commerce platform and preprocess the product information; obtain product review information from the historical transaction records of the e-commerce platform and calculate the degree of difference between the product information and the product review information;

[0006] Step S2: Based on the calculated degree of difference, obtain the target product; track the product information modification records of the target product and calculate the information modification rate; combine the complaint rate and return rate of the target product in historical transactions to calculate the risk warning coefficient;

[0007] Step S3: Based on the calculated risk warning coefficient and the degree of difference, classify the target products into different levels; implement risk response strategies for target products of different levels.

[0008] Step S4: Monitor the response effect of the risk response strategy in real time and calculate the response effect coefficient; dynamically adjust the risk response strategy based on the calculated response effect coefficient.

[0009] Furthermore, step S1 includes:

[0010] Step S1-1: Obtain product information for all goods in real time through the data interface of the e-commerce platform; the product information includes product text data, product image data, and product video data; collect product review information from user feedback in historical transaction records; the product review information includes actual product text data, actual product image data, and actual product video data;

[0011] Step S1-2: Extract text feature vectors from product text data and image feature vectors from product image data using the model; process the video frame sequence of each product video segment in the product video data to extract dynamic feature vectors; extract actual text feature vectors, actual image feature vectors, and actual dynamic feature vectors from the product review information; calculate the degree of text difference, image difference, and video difference respectively.

[0012] The degree of textual difference is:

[0013] ;

[0014] Among them, D t T represents the degree of textual difference. a T represents the text feature vector. b This represents the actual text feature vector;

[0015] The degree of difference in the images is:

[0016] ;

[0017] Among them, D p P represents the image difference. a P represents the image feature vector. b Represents the actual image feature vector;

[0018] The degree of difference in the videos is as follows:

[0019] ;

[0020] Among them, D m Indicating the degree of difference in videos, M a M represents the video feature vector. b Represents the actual video vector;

[0021] Step S1-3: Calculate the degree of difference between product information published on the e-commerce platform and product review information:

[0022] ;

[0023] Where D represents the degree of difference between product information and product review information published on the e-commerce platform, and α, β, and γ are all weighting coefficients;

[0024] The aforementioned "actual product text data" includes user reviews of products, text descriptions of return reasons, and text explanations in complaints. This data is transformed into structured actual product text data through data cleaning and preprocessing. "Actual product image data" refers to photos of the actual product uploaded by users, and "actual product video data" refers to videos of the actual product taken by users. In the above steps, for text data, natural language processing models such as BERT and WordVec are used to extract semantic vectors from product titles and detailed descriptions. For image data, convolutional neural networks such as ResNet and VGG are used to extract features such as color, texture, and shape. For video data, dynamic features are extracted through 3D neural networks or frame sequence processing. All extracted features are ultimately transformed into numerical vectors of uniform dimension, ensuring that differences can be calculated using cosine similarity.

[0025] Furthermore, step S2 includes:

[0026] Step S2-1: Identify the product categories of all goods published on the e-commerce platform, divide all goods according to the identified product categories, and establish a difference degree benchmark library; calculate the average difference degree of each product category. and standard deviation Generate the threshold for the degree of difference for each product category:

[0027] ;

[0028] in, This is an adjustment coefficient; when the degree of difference of a certain product D ≥ D th The product is marked as the target product;

[0029] Step S2-2: Monitor the product information of the target product, count the number of times the product information of the target product is modified within the target time period, and extract text modification data, image modification data, and video modification data from each product information modification record; associate the production batch information of the target product with the preset product traceability code, and determine whether the information modification corresponds to product review information; if the information modification is consistent with the product review information, calculate the modification rationality coefficient:

[0030] ;

[0031] Among them, H i This represents the reasonableness coefficient of the modification of the i-th modification record. Indicates the degree of difference in text features before and after modification. This indicates the degree of difference in image features before and after the modification. This indicates the degree of difference in video features before and after the modification. , , The difference weights are respectively for text, images, and videos;

[0032] Step S2-3: Set the threshold H for modifying the rationality coefficient. th When H i ≥H th When H is active, it is marked as a valid modification; when H is active, it is marked as a valid modification. i <H th If the modification is invalid, it is marked as invalid. If the product information of the target product is invalid or inconsistent with the product review information, then based on the text modification data, image modification data, and video modification data, text feature change, image feature change, and video feature change are generated; the information modification rate of the target product is calculated.

[0033] ;

[0034] Where R represents the information modification rate, ∆D ti ∆D represents the change in text features in the i-th product information modification record. pi ∆D represents the change in image features in the i-th product information modification record. mi This represents the amount of change in video features in the i-th product information modification record. As time weight, Indicates the target time period;

[0035] Step S2-4: Collect the total number of completed orders, the number of complaint orders, and the number of returned orders for the target product in historical transactions, and calculate the complaint rate and return rate of the target product;

[0036] The complaint rate:

[0037] ;

[0038] Where C represents the complaint rate, m represents the number of complaint orders for the target product in historical transactions, and Q represents the total number of completed orders for the target product in historical transactions;

[0039] The return rate:

[0040] ;

[0041] Where S represents the return rate, and n represents the number of returned orders for the target product in historical transactions;

[0042] Step S2-5: Calculate the risk warning coefficient for the target product based on its information modification rate, complaint rate, and return rate.

[0043] ;

[0044] Where K represents the risk warning coefficient of the target product. , , All are weighting coefficients;

[0045] In the above steps, the identification method for "product category" can be implemented through two layers of logic: the basic layer: preliminary classification based on the category labels filled in by the merchant (such as "women's clothing - shirts"); the verification layer: semantic analysis of product titles and descriptions using a text classification model (such as TextCNN) to correct incorrectly filled categories.

[0046] Furthermore, step S3 includes:

[0047] Step S3-1: Obtain all product information in the risk event records of the e-commerce platform, divide the product information according to product category, extract all historical product information of the same product category as the target product, calculate the risk warning coefficient and degree of difference based on the historical product information, and divide the risk coefficient range and degree of difference range according to the set risk coefficient threshold and degree of difference threshold.

[0048] Step S3-2: Using the risk warning coefficient and the degree of difference as two-dimensional classification indicators, establish a risk level classification matrix; the horizontal axis of the risk level classification matrix is ​​the risk warning coefficient range, and the vertical axis is the degree of difference range. By cross-combining the horizontal risk warning coefficient and the vertical degree of difference, the risk level of the target commodity is determined.

[0049] Step S3-3: For target products with different risk levels, match a preset risk response strategy library, which includes dynamic response rules. The dynamic response rules include: for high-risk products, triggering an immediate removal verification mechanism, simultaneously initiating a credit rating deduction process, and pushing risk warnings to users of the e-commerce platform; for medium-risk products, implementing mandatory labeling and correction measures for product information, restricting medium-risk products from participating in promotional activities on the e-commerce platform, and requiring the submission of a product qualification re-inspection report within a preset time; for low-risk products, sending a risk warning to the backend, prompting merchants to independently rectify low-risk products, and only products that have passed re-inspection after rectification can have their warning status lifted; and storing the risk level classification results in association with the risk response strategy execution records.

[0050] In the above steps, the scope of "risk event records" includes not only information on products that have been judged as risky in the past, but also disputed orders caused by inconsistent product information, user complaint records, platform penalty cases, etc., to ensure that the extracted historical data can comprehensively reflect the risk characteristics.

[0051] Furthermore, step S4 includes:

[0052] Step S4-1: Construct a multi-dimensional evaluation index system for response effectiveness, which includes risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate.

[0053] The risk mitigation rate is calculated by statistically analyzing the number of new complaint orders and new return orders for the target product within a preset monitoring period after the implementation of the risk response strategy.

[0054] ;

[0055] Where E1 represents the risk mitigation rate, Q 前 Q represents the number of complaint orders and return orders within the same monitoring period before the implementation of the risk response strategy. 后 This indicates the number of complaint orders and return orders within the preset monitoring period after the risk response strategy is implemented;

[0056] The user satisfaction improvement rate is calculated by collecting user review data of the target product through the user feedback system of the e-commerce platform, extracting sentiment characteristics from the review text, and then calculating the user satisfaction improvement rate.

[0057] ;

[0058] Where E2 represents the user satisfaction improvement rate, S 前 S represents the average user satisfaction score before the implementation of the risk response strategy. 后 This represents the average user satisfaction score after the implementation of the risk response strategy;

[0059] The timeliness compliance rate is calculated by recording the time from receiving the response strategy instruction to completing the rectification and passing the re-inspection by the e-commerce platform.

[0060] ;

[0061] Among them, T 实 ≤T 标 When T 实 >T 标 At that time, E3=0; E3 represents the timeliness compliance rate, T 标 T represents the standard rectification period for the target product. 实 Indicates the actual rectification time;

[0062] Step S4-2: Determine the weight coefficients of each evaluation indicator using the analytic hierarchy process (AHP) and calculate the response effect coefficients.

[0063] ;

[0064] in, , , These are the weighting coefficients for risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate, respectively.

[0065] Step S4-3: Establish a dynamic adjustment model for risk response strategies and optimize the risk response strategies based on the response effect coefficients.

[0066] Furthermore, to better implement the above methods, an AI-based platform risk event intelligent response system is also provided. This system includes: an information collection and difference analysis module, a risk warning coefficient calculation module, a risk classification and response module, and an effect evaluation and strategy adjustment module.

[0067] The information collection and difference analysis module is used to collect product information published on e-commerce platforms and product review information from historical transactions, and calculate the degree of difference after preprocessing this information;

[0068] The risk warning coefficient calculation module is used to filter out target products with excessive differences, track the information modification records of the target products and calculate the information modification rate, and calculate the risk warning coefficient by combining the complaint rate and return rate in historical transactions.

[0069] The risk classification and response module classifies target products into different levels based on risk warning coefficients and degree of difference, and implements corresponding risk response strategies for products of different levels.

[0070] The effect evaluation and strategy adjustment module evaluates the implementation effect of risk response strategies from multiple dimensions, calculates the response effect coefficient, and dynamically adjusts the risk response strategies based on the response effect coefficient.

[0071] Furthermore, the information acquisition and difference analysis module includes: an information acquisition unit, a feature extraction unit, and a difference calculation unit;

[0072] The information collection unit obtains product information published on the e-commerce platform and product review information from historical transaction records in real time through the data interface of the e-commerce platform;

[0073] The feature extraction unit is used to extract text feature vectors from product text data, image feature vectors from product image data, and dynamic feature vectors from video frame sequences of product video data; at the same time, it extracts the corresponding actual feature vectors from product review information.

[0074] The difference calculation unit calculates the degree of difference between text, image, and video respectively, and combines the weighting coefficients to calculate the overall degree of difference between product information and product review information.

[0075] Furthermore, the risk warning coefficient calculation module includes: a target product identification unit, a modification rate calculation unit, a complaint return rate calculation unit, and a warning coefficient unit;

[0076] The target product identification unit divides products according to product categories and establishes a difference degree benchmark library. It calculates the mean and standard deviation of the difference degree for each product category, generates a difference degree threshold, and marks products with a difference degree exceeding the threshold as target products.

[0077] The modification rate calculation unit is used to monitor the information modification records of the target product within the target time period, extract various types of modification data and generate feature variables, and calculate the information modification rate by combining time weights.

[0078] The complaint and return rate calculation unit is used to collect the total number of completed orders, the number of complaint orders, and the number of return orders in the historical transactions of the target product, and to calculate the complaint rate and return rate respectively.

[0079] The early warning coefficient unit calculates the risk early warning coefficient based on the information modification rate, complaint rate, and return rate, combined with the corresponding weighting coefficients.

[0080] Furthermore, the risk classification and response module includes: an interval division unit, a level determination unit, and a strategy execution unit;

[0081] The interval division unit extracts historical product information belonging to the same category as the target product, calculates the risk warning coefficient and degree of difference of the historical product information, and divides the risk warning coefficient interval and degree of difference interval according to the set threshold value.

[0082] The risk level determination unit establishes a risk level classification matrix using risk warning coefficient and degree of difference as two-dimensional indicators. The risk level of the target commodity is determined by the cross combination of horizontal risk warning coefficient and vertical degree of difference.

[0083] The strategy execution unit matches and executes dynamic response rules from the preset risk response strategy library for products with different risk levels, while also storing the risk level classification results and response strategy execution records in association.

[0084] Furthermore, the effect evaluation and strategy adjustment module includes: evaluation indicator unit, effect coefficient unit, and strategy adjustment unit;

[0085] The evaluation indicator unit constructs a multi-dimensional evaluation indicator system for response effectiveness, including risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate.

[0086] The effect coefficient unit uses the analytic hierarchy process (AHP) to determine the weight coefficients of each evaluation indicator and then calculates the response effect coefficient by combining the numerical values ​​of each indicator.

[0087] The strategy adjustment unit establishes a dynamic adjustment model for risk response strategies and optimizes and adjusts risk response strategies based on response effectiveness coefficients.

[0088] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention extracts features from multiple dimensions to calculate the degree of difference between product information and product review information published on e-commerce platforms. Combined with a difference degree benchmark library, it accurately filters out target products whose product information and product review information do not match. By establishing a two-dimensional level matrix based on risk warning coefficients and difference degrees, risk response strategies are implemented for different risk levels, balancing risk control with the flexibility of merchant operations. The response results are evaluated through multiple indicators such as risk mitigation rate and user satisfaction improvement. Combined with a dynamic adjustment model, the strategy is continuously optimized to improve the timeliness and effectiveness of risk handling, enhance user trust, and ensure the sustainable development of the platform. Attached Figure Description

[0089] Figure 1 This is a schematic diagram of the method flow of an intelligent response system and method for platform risk events based on artificial intelligence according to the present invention;

[0090] Figure 2 This is a schematic diagram of the system structure of an intelligent response system and method for platform risk events based on artificial intelligence, according to the present invention. Detailed Implementation

[0091] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0092] Example 1: As Figure 1 As shown, the present invention provides a technical solution, an intelligent response method for platform risk events based on artificial intelligence, the intelligent response method comprising:

[0093] Step S1: Collect product information published by all merchants on the e-commerce platform and preprocess the product information; obtain product review information published by all buyers in the historical transaction records of the e-commerce platform, and calculate the degree of difference between the product information and the product review information;

[0094] Step S1 includes:

[0095] Step S1-1: Obtain product information for all goods in real time through the data interface of the e-commerce platform; the product information includes product text data, product image data, and product video data; collect product review information from user feedback in historical transaction records; the product review information includes actual product text data, actual product image data, and actual product video data;

[0096] Step S1-2: Extract text feature vectors from product text data and image feature vectors from product image data using the model; process the video frame sequence of each product video segment in the product video data to extract dynamic feature vectors; extract actual text feature vectors, actual image feature vectors, and actual dynamic feature vectors from the product review information; calculate the degree of text difference, image difference, and video difference respectively.

[0097] The degree of textual difference is:

[0098] ;

[0099] Among them, D t T represents the degree of textual difference. a T represents the text feature vector. b This represents the actual text feature vector;

[0100] The degree of difference in the images is:

[0101] ;

[0102] Among them, D p P represents the image difference. a P represents the image feature vector. b Represents the actual image feature vector;

[0103] The degree of difference in the videos is as follows:

[0104] ;

[0105] Among them, D m Indicating the degree of difference in videos, M a M represents the video feature vector. b Represents the actual video vector;

[0106] Step S1-3: Calculate the degree of difference between product information published on the e-commerce platform and product review information:

[0107] ;

[0108] Where D represents the degree of difference between product information and product review information published on the e-commerce platform, and α, β, and γ are all weighting coefficients;

[0109] Step S2: Based on the calculated degree of difference, obtain the target product; track the product information modification records of the target product and calculate the information modification rate; combine the complaint rate and return rate of the target product in historical transactions to calculate the risk warning coefficient;

[0110] Step S2 includes:

[0111] Step S2-1: Identify the product categories of all goods published on the e-commerce platform, divide all goods according to the identified product categories, and establish a difference degree benchmark library; calculate the average difference degree of each product category. and standard deviation Generate the threshold for the degree of difference for each product category:

[0112] ;

[0113] in, This is an adjustment coefficient; when the degree of difference of a certain product D ≥ D th The product is marked as the target product;

[0114] Step S2-2: Monitor the product information of the target product, count the number of times the product information of the target product is modified within the target time period, and extract text modification data, image modification data, and video modification data from each product information modification record; associate the production batch information of the target product with the preset product traceability code, and determine whether the information modification corresponds to product review information; if the information modification is consistent with the product review information, calculate the modification rationality coefficient:

[0115] ;

[0116] Among them, H i This represents the reasonableness coefficient of the modification of the i-th modification record. Indicates the degree of difference in text features before and after modification. This indicates the degree of difference in image features before and after the modification. This indicates the degree of difference in video features before and after the modification. , , The difference weights are respectively for text, images, and videos;

[0117] Step S2-3: Set the threshold H for modifying the rationality coefficient. th When H i ≥H thWhen H is active, it is marked as a valid modification; when H is active, it is marked as a valid modification. i <H th If the modification is invalid, it is marked as invalid. If the product information of the target product is invalid or inconsistent with the product review information, then based on the text modification data, image modification data, and video modification data, text feature change, image feature change, and video feature change are generated; the information modification rate of the target product is calculated.

[0118] ;

[0119] Where R represents the information modification rate, ∆D ti ∆D represents the change in text features in the i-th product information modification record. pi ∆D represents the change in image features in the i-th product information modification record. mi This represents the amount of change in video features in the i-th product information modification record. As time weight, Indicates the target time period;

[0120] Step S2-4: Collect the total number of completed orders, the number of complaint orders, and the number of returned orders for the target product in historical transactions, and calculate the complaint rate and return rate of the target product;

[0121] The complaint rate:

[0122] ;

[0123] Where C represents the complaint rate, m represents the number of complaint orders for the target product in historical transactions, and Q represents the total number of completed orders for the target product in historical transactions;

[0124] The return rate:

[0125] ;

[0126] Where S represents the return rate, and n represents the number of returned orders for the target product in historical transactions;

[0127] Step S2-5: Calculate the risk warning coefficient for the target product based on its information modification rate, complaint rate, and return rate.

[0128] ;

[0129] Where K represents the risk warning coefficient of the target product. , , All are weighting coefficients;

[0130] Step S3: Based on the calculated risk warning coefficient and the degree of difference, classify the target products into different levels; implement risk response strategies for target products of different levels.

[0131] Step S3 includes:

[0132] Step S3-1: Obtain all product information in the risk event records of the e-commerce platform, divide the product information according to product category, extract all historical product information of the same product category as the target product, calculate the risk warning coefficient and degree of difference based on the historical product information, and divide the risk coefficient range and degree of difference range according to the set risk coefficient threshold and degree of difference threshold.

[0133] Step S3-2: Using the risk warning coefficient and the degree of difference as two-dimensional classification indicators, establish a risk level classification matrix; the horizontal axis of the risk level classification matrix is ​​the risk warning coefficient range, and the vertical axis is the degree of difference range. By cross-combining the horizontal risk warning coefficient and the vertical degree of difference, the risk level of the target commodity is determined.

[0134] Step S3-3: For target products with different risk levels, match a preset risk response strategy library, which includes dynamic response rules. The dynamic response rules include: for high-risk products, triggering an immediate removal verification mechanism, simultaneously initiating a credit rating deduction process, and pushing risk warnings to users of the e-commerce platform; for medium-risk products, implementing mandatory labeling and correction measures for product information, restricting medium-risk products from participating in promotional activities on the e-commerce platform, and requiring the submission of a product qualification re-inspection report within a preset time; for low-risk products, sending a risk warning to the backend, prompting merchants to independently rectify low-risk products, and only products that have passed re-inspection after rectification can have their warning status lifted; and storing the risk level classification results in association with the risk response strategy execution records.

[0135] Step S4: Monitor the response effect of the risk response strategy in real time and calculate the response effect coefficient; dynamically adjust the risk response strategy based on the calculated response effect coefficient;

[0136] Step S4 includes:

[0137] Step S4-1: Construct a multi-dimensional evaluation index system for response effectiveness, which includes risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate.

[0138] The risk mitigation rate is calculated by statistically analyzing the number of new complaint orders and new return orders for the target product within a preset monitoring period after the implementation of the risk response strategy.

[0139] ;

[0140] Where E1 represents the risk mitigation rate, Q 前 Q represents the number of complaint orders and return orders within the same monitoring period before the implementation of the risk response strategy. 后 This indicates the number of complaint orders and return orders within the preset monitoring period after the risk response strategy is implemented;

[0141] The user satisfaction improvement rate is calculated by collecting user review data of the target product through the user feedback system of the e-commerce platform, extracting sentiment characteristics from the review text, and then calculating the user satisfaction improvement rate.

[0142] ;

[0143] Where E2 represents the user satisfaction improvement rate, S 前 S represents the average user satisfaction score before the implementation of the risk response strategy. 后 This represents the average user satisfaction score after the implementation of the risk response strategy;

[0144] The timeliness compliance rate is calculated by recording the time from receiving the response strategy instruction to completing the rectification and passing the re-inspection by the e-commerce platform.

[0145] ;

[0146] Among them, T 实 ≤T 标 When T 实 >T 标 At that time, E3=0; E3 represents the timeliness compliance rate, T 标 T represents the standard rectification period for the target product. 实 Indicates the actual rectification time;

[0147] Step S4-2: Determine the weight coefficients of each evaluation indicator using the analytic hierarchy process (AHP) and calculate the response effect coefficients.

[0148] ;

[0149] in, , , These are the weighting coefficients for risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate, respectively.

[0150] Step S4-3: Establish a dynamic adjustment model for risk response strategies and optimize the risk response strategies based on the response effectiveness coefficients;

[0151] Example 2: Figure 2As shown, in order to better implement the above method, an intelligent response system for platform risk events based on artificial intelligence is also provided. The intelligent response system includes: an information collection and difference analysis module, a risk warning coefficient calculation module, a risk classification and response module, and an effect evaluation and strategy adjustment module.

[0152] The information collection and difference analysis module is used to collect product information published on e-commerce platforms and product review information from historical transactions, and calculate the degree of difference after preprocessing this information;

[0153] The risk warning coefficient calculation module is used to filter out target products with excessive differences, track the information modification records of the target products and calculate the information modification rate, and calculate the risk warning coefficient by combining the complaint rate and return rate in historical transactions.

[0154] The risk classification and response module classifies target products into different levels based on risk warning coefficients and degree of difference, and implements corresponding risk response strategies for products of different levels.

[0155] The effect evaluation and strategy adjustment module evaluates the execution effect of risk response strategies from multiple dimensions, calculates the response effect coefficient, and dynamically adjusts the risk response strategies based on the response effect coefficient.

[0156] The information collection and difference analysis module includes: an information collection unit, a feature extraction unit, and a difference calculation unit.

[0157] The information collection unit obtains product information published on the e-commerce platform and product review information from historical transaction records in real time through the data interface of the e-commerce platform;

[0158] The feature extraction unit is used to extract text feature vectors from product text data, image feature vectors from product image data, and dynamic feature vectors from video frame sequences of product video data; at the same time, it extracts the corresponding actual feature vectors from product review information.

[0159] The difference calculation unit calculates the degree of difference between text, image, and video respectively, and combines the weighting coefficients to calculate the overall degree of difference between product information and product review information.

[0160] The risk warning coefficient calculation module includes: target product identification unit, modification rate calculation unit, complaint return rate calculation unit, and warning coefficient unit.

[0161] The target product identification unit divides products according to product categories and establishes a difference degree benchmark library. It calculates the mean and standard deviation of the difference degree for each product category, generates a difference degree threshold, and marks products with a difference degree exceeding the threshold as target products.

[0162] The modification rate calculation unit is used to monitor the information modification records of the target product within the target time period, extract various types of modification data and generate feature variables, and calculate the information modification rate by combining time weights.

[0163] The complaint and return rate calculation unit is used to collect the total number of completed orders, the number of complaint orders, and the number of return orders in the historical transactions of the target product, and to calculate the complaint rate and return rate respectively.

[0164] The early warning coefficient unit calculates the risk early warning coefficient based on the information modification rate, complaint rate, and return rate, combined with the corresponding weighting coefficients.

[0165] The risk classification and response module includes: an interval division unit, a level determination unit, and a strategy execution unit.

[0166] The interval division unit extracts historical product information belonging to the same category as the target product, calculates the risk warning coefficient and degree of difference of the historical product information, and divides the risk warning coefficient interval and degree of difference interval according to the set threshold value.

[0167] The risk level determination unit establishes a risk level classification matrix using risk warning coefficient and degree of difference as two-dimensional indicators. The risk level of the target commodity is determined by the cross combination of horizontal risk warning coefficient and vertical degree of difference.

[0168] The strategy execution unit matches and executes dynamic response rules from the preset risk response strategy library for products with different risk levels, and at the same time associates and stores the risk level classification results with the response strategy execution records.

[0169] The effect evaluation and strategy adjustment module includes: evaluation index unit, effect coefficient unit, and strategy adjustment unit.

[0170] The evaluation indicator unit constructs a multi-dimensional evaluation indicator system for response effectiveness, including risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate.

[0171] The effect coefficient unit uses the analytic hierarchy process (AHP) to determine the weight coefficients of each evaluation indicator and then calculates the response effect coefficient by combining the numerical values ​​of each indicator.

[0172] The strategy adjustment unit establishes a dynamic adjustment model for risk response strategies and optimizes and adjusts risk response strategies based on response effectiveness coefficients.

[0173] In an embodiment of the present invention, taking a "pure cotton shirt" product on a certain platform as an example:

[0174] The system obtains shirt information posted by merchants through the platform interface (text description "100% pure cotton", display image of a smooth white shirt, video demonstration of soft fabric), and at the same time extracts actual information from buyer feedback in historical transactions (actual text label "contains 30% chemical fiber", buyer-uploaded photos of actual products with obvious pilling, video showing that the fabric is relatively stiff).

[0175] The degree of text difference D is obtained by feature vector extraction and cosine similarity calculation. t =0.6, Image Difference Level D p =0.5, Video Difference Level D m =0.4, and combining the weights α=0.4, β=0.3, and γ=0.3, the total degree of difference D=0.4×0.6+0.3×0.5+0.3×0.4=0.51;

[0176] The average degree of difference in the "men's tops" category described for this shirt D =0.2, standard deviation D =0.15, adjustment factor =2, threshold D th =0.2 + 2 × 0.15 = 0.5; Since D = 0.51 > 0.5, mark it as the target product;

[0177] The information modification records of the target product within 30 days were tracked, with a total of 2 modifications (text changed from "pure cotton" to "cotton blend", image replaced once), and the information modification rate was calculated as R=0.3; in historical transactions, the total number of orders was Q=200, complaints were m=30, and returns were n=40, resulting in a complaint rate C=0.15 and a return rate S=0.2; combined with weights... 1 = 0.2 2 = 0.4 =3=0.4, risk warning coefficient K=0.2×0.3+0.4×0.15+0.4×0.2=0.18;

[0178] Historical data for the "Men's Tops" category shows that a risk warning coefficient > 0.15 and a difference degree > 0.5 indicate a medium risk. Therefore, this shirt is classified as a medium risk, and the response strategy is as follows: mandatory labeling "contains 30% synthetic fiber", restriction on participation in platform promotions, and a requirement for the merchant to submit a fabric testing report within 3 days.

[0179] 30-day monitoring after implementation: The original cycle of 70 complaints and returns has been reduced to 20, with a risk mitigation rate of E1=(70-20) / 70=0.71; user satisfaction has increased from 3.2 to 4.5, with an improvement rate of E2=(4.5-3.2) / 3.2=0.41; the merchant completed rectification in 2 days, with a timeliness compliance rate of E3=1; the response effect coefficient is qualified, and the strategy remains unchanged.

[0180] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent response method for platform risk events based on artificial intelligence, characterized in that: The intelligent response method includes: Step S1: Collect product information published on the e-commerce platform and preprocess the product information; obtain product review information from the historical transaction records of the e-commerce platform and calculate the degree of difference between the product information and the product review information; Step S2: Based on the calculated degree of difference, obtain the target product; track the product information modification records of the target product and calculate the information modification rate; combine the complaint rate and return rate of the target product in historical transactions to calculate the risk warning coefficient; Step S3: Based on the calculated risk warning coefficient and the degree of difference, classify the target products into different levels; implement risk response strategies for target products of different levels. Step S4: Monitor the response effect of the risk response strategy in real time and calculate the response effect coefficient; dynamically adjust the risk response strategy based on the calculated response effect coefficient; Step S1 includes: Step S1-1: Obtain product information for all goods in real time through the data interface of the e-commerce platform; the product information includes product text data, product image data, and product video data; collect product review information from user feedback in historical transaction records; the product review information includes actual product text data, actual product image data, and actual product video data; Step S1-2: Extract text feature vectors from product text data and image feature vectors from product image data using the model; process the video frame sequence of each product video segment in the product video data to extract dynamic feature vectors; extract actual text feature vectors, actual image feature vectors, and actual dynamic feature vectors from the product review information; calculate the degree of text difference, image difference, and video difference respectively. The degree of textual difference is: ； Among them, D t T represents the degree of textual difference. a T represents the text feature vector. b This represents the actual text feature vector; The degree of difference in the images is: ； Among them, D p P represents the image difference. a P represents the image feature vector. b Represents the actual image feature vector; The degree of difference in the videos is as follows: ； Among them, D m Indicating the degree of difference in videos, M a M represents the video feature vector. b Represents the actual video vector; Step S1-3: Calculate the degree of difference between product information published on the e-commerce platform and product review information: ； Where D represents the degree of difference between product information and product review information published on the e-commerce platform, and α, β, and γ are all weighting coefficients; Step S2 includes: Step S2-1: Identify the product categories of all goods published on the e-commerce platform, divide all goods according to the identified product categories, and establish a difference degree benchmark library; calculate the average difference degree of each product category. and standard deviation Generate the threshold for the degree of difference for each product category: ； in, This is an adjustment coefficient; when the degree of difference of a certain product D ≥ D th The product is marked as the target product; Step S2-2: Monitor the product information of the target product, count the number of times the product information of the target product is modified within the target time period, and extract text modification data, image modification data, and video modification data from each product information modification record; associate the production batch information of the target product with the preset product traceability code, and determine whether the information modification corresponds to product review information; if the information modification is consistent with the product review information, calculate the modification rationality coefficient: ； Among them, H i This represents the reasonableness coefficient of the modification of the i-th modification record. Indicates the degree of difference in text features before and after modification. This indicates the degree of difference in image features before and after the modification. This indicates the degree of difference in video features before and after the modification. , , The difference weights are respectively for text, image, and video; Step S2-3: Set the threshold H for modifying the rationality coefficient. th When H i ≥H th When H is active, it is marked as a valid modification; when H is active, it is marked as a valid modification. i <H th If the modification is invalid, it is marked as invalid. If the product information of the target product is invalid or inconsistent with the product review information, then based on the text modification data, image modification data, and video modification data, text feature change, image feature change, and video feature change are generated; the information modification rate of the target product is calculated. ； Where R represents the information modification rate, ∆D ti ∆D represents the change in text features in the i-th product information modification record. pi ∆D represents the change in image features in the i-th product information modification record. mi This represents the amount of change in video features in the i-th product information modification record. As time weight, Indicates the target time period; Step S2-4: Collect the total number of completed orders, the number of complaint orders, and the number of returned orders for the target product in historical transactions, and calculate the complaint rate and return rate of the target product; The complaint rate: ； Where C represents the complaint rate, m represents the number of complaint orders for the target product in historical transactions, and Q represents the total number of completed orders for the target product in historical transactions; The return rate: ； Where S represents the return rate, and n represents the number of returned orders for the target product in historical transactions; Step S2-5: Calculate the risk warning coefficient for the target product based on its information modification rate, complaint rate, and return rate. ； Where K represents the risk warning coefficient of the target product. , , All are weighting coefficients.

2. The intelligent response method for platform risk events based on artificial intelligence according to claim 1, characterized in that: Step S3 includes: Step S3-1: Obtain all product information in the risk event records of the e-commerce platform, divide the product information according to product category, extract all historical product information of the same product category as the target product, calculate the risk warning coefficient and degree of difference based on the historical product information, and divide the risk coefficient range and degree of difference range according to the set risk coefficient threshold and degree of difference threshold. Step S3-2: Using the risk warning coefficient and the degree of difference as two-dimensional classification indicators, establish a risk level classification matrix; the horizontal axis of the risk level classification matrix is ​​the risk warning coefficient range, and the vertical axis is the degree of difference range. By cross-combining the horizontal risk warning coefficient and the vertical degree of difference, the risk level of the target commodity is determined. Step S3-3: For target products with different risk levels, match a preset risk response strategy library, which includes dynamic response rules. The dynamic response rules include: for high-risk products, trigger an immediate removal verification mechanism, simultaneously initiate a credit rating deduction process, and push risk warnings to users of the e-commerce platform; for medium-risk products, implement mandatory labeling and correction measures for product information, restrict medium-risk products from participating in promotional activities on the e-commerce platform, and require the submission of a product qualification re-inspection report within a preset time; for low-risk products, send a risk warning to the backend, prompting the low-risk products to conduct self-rectification, and the warning status can be lifted only after the rectified products pass a re-inspection; and associate and store the risk level classification results with the risk response strategy execution records.

3. The intelligent response method for platform risk events based on artificial intelligence according to claim 1, characterized in that: Step S4 includes: Step S4-1: Construct a multi-dimensional evaluation index system for response effectiveness, which includes risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate. The risk mitigation rate is calculated by statistically analyzing the number of new complaint orders and new return orders for the target product within a preset monitoring period after the implementation of the risk response strategy. ； Where E1 represents the risk mitigation rate, Q 前 Q represents the number of complaint orders and return orders within the same monitoring period before the implementation of the risk response strategy. 后 This indicates the number of complaint orders and return orders within the preset monitoring period after the risk response strategy is implemented; The user satisfaction improvement rate is calculated by collecting user review data of the target product through the user feedback system of the e-commerce platform, extracting sentiment characteristics from the review text, and then calculating the user satisfaction improvement rate. ； Where E2 represents the user satisfaction improvement rate, S 前 S represents the average user satisfaction score before the implementation of the risk response strategy. 后 This represents the average user satisfaction score after the implementation of the risk response strategy; The timeliness compliance rate is calculated by recording the time from receiving the response strategy instruction to completing the rectification and passing the re-inspection by the e-commerce platform. ； Among them, T 实 ≤T 标 When T 实 >T 标 At that time, E3=0; E3 represents the timeliness compliance rate, T 标 T represents the standard rectification period for the target product. 实 Indicates the actual rectification time; Step S4-2: Determine the weight coefficients of each evaluation indicator using the analytic hierarchy process (AHP) and calculate the response effect coefficients. ； in, , , These are the weighting coefficients for risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate, respectively. Step S4-3: Establish a dynamic adjustment model for risk response strategies and optimize the risk response strategies based on the response effect coefficients.

4. An AI-based intelligent response system for platform risk events, used to execute the AI-based intelligent response method for platform risk events as described in any one of claims 1-3, characterized in that: The intelligent response system includes: an information collection and difference analysis module, a risk warning coefficient calculation module, a risk classification and response module, and an effect evaluation and strategy adjustment module; The information collection and difference analysis module is used to collect product information published on the e-commerce platform and product review information from historical transactions, and calculate the degree of difference after preprocessing this information; The risk warning coefficient calculation module is used to filter out target products with excessive differences, track the information modification records of the target products and calculate the information modification rate, and calculate the risk warning coefficient by combining the complaint rate and return rate in historical transactions. The risk classification and response module classifies target products into different levels based on risk warning coefficients and degree of difference, and executes corresponding risk response strategies for products of different levels. The effect evaluation and strategy adjustment module evaluates the execution effect of the risk response strategy from multiple dimensions, calculates the response effect coefficient, and dynamically adjusts the risk response strategy based on the response effect coefficient.

5. The platform risk event intelligent response system based on artificial intelligence according to claim 4, characterized in that: The information acquisition and difference analysis module includes: an information acquisition unit, a feature extraction unit, and a difference calculation unit; The information collection unit obtains product information published on the e-commerce platform and product review information from historical transaction records in real time through the data interface of the e-commerce platform. The feature extraction unit is used to extract text feature vectors from product text data, extract image feature vectors from product image data, and extract dynamic feature vectors after processing the video frame sequence of product video data; at the same time, it extracts the corresponding actual feature vectors from product review information. The difference calculation unit calculates the degree of text difference, image difference, and video difference respectively, and calculates the overall degree of difference between product information and product review information by combining the weighting coefficients.

6. The platform risk event intelligent response system based on artificial intelligence according to claim 4, characterized in that: The risk warning coefficient calculation module includes: a target product identification unit, a modification rate calculation unit, a complaint return rate calculation unit, and a warning coefficient unit; The target product identification unit divides products according to product categories and establishes a difference degree benchmark library, calculates the mean and standard deviation of the difference degree of each product category, generates a difference degree threshold, and marks products with a difference degree exceeding the threshold as target products. The modification rate calculation unit is used to monitor the information modification records of the target product within the target time period, extract various types of modification data and generate feature variables, and calculate the information modification rate by combining time weights. The complaint and return rate calculation unit is used to collect the total number of completed orders, the number of complaint orders, and the number of return orders in the historical transactions of the target product, and to calculate the complaint rate and return rate respectively. The warning coefficient unit calculates the risk warning coefficient based on the information modification rate, complaint rate, and return rate, combined with the corresponding weighting coefficients.

7. The platform risk event intelligent response system based on artificial intelligence according to claim 4, characterized in that: The risk classification and response module includes: an interval division unit, a level determination unit, and a strategy execution unit; The interval division unit extracts historical product information belonging to the same category as the target product, calculates the risk warning coefficient and degree of difference of the historical product information, and divides the risk warning coefficient interval and degree of difference interval according to the set threshold value. The risk level determination unit establishes a risk level classification matrix using risk warning coefficient and degree of difference as two-dimensional indicators. By cross-combining the horizontal risk warning coefficient and the vertical degree of difference, the risk level of the target commodity is determined. The strategy execution unit matches and executes dynamic response rules from a preset risk response strategy library for products with different risk levels, and simultaneously associates and stores the risk level classification results with the response strategy execution records.

8. The platform risk event intelligent response system based on artificial intelligence according to claim 4, characterized in that: The effect evaluation and strategy adjustment module includes: an evaluation index unit, an effect coefficient unit, and a strategy adjustment unit; The evaluation indicator unit constructs a multi-dimensional evaluation indicator system for response effectiveness, including risk mitigation rate, user satisfaction improvement rate, and timeliness achievement rate. The effect coefficient unit determines the weight coefficient of each evaluation index through the analytic hierarchy process, and calculates the response effect coefficient by combining the numerical values ​​of each index. The strategy adjustment unit establishes a dynamic adjustment model for risk response strategies and optimizes and adjusts the risk response strategies based on the response effect coefficient.

Images